A comprehensive review on accuracy

in ultrasonic flow measurement using

reconfigurable systems and deep learning
approaches
Cite as: AIP Advances 10, 105221 (2020); https://doi.org/10.1063/5.0022154
Submitted: 25 July 2020 • Accepted: 16 September 2020 • Published Online: 13 October 2020

Senthil Kumar J, Kamaraj A, Kalyana Sundaram C, et al.

ARTICLES YOU MAY BE INTERESTED IN

Review of ultrasonic measurement methods for two-phase flow

Review of Scientific Instruments 92, 091502 (2021); https://doi.org/10.1063/5.0049046

Signal processing method based on connection fitting of echo peak point with a large slope
for ultrasonic gas flow meter
Review of Scientific Instruments 92, 015105 (2021); https://doi.org/10.1063/5.0021801

Time-of-flight dependency on transducer separation distance in a reflective-path guided-

wave ultrasonic flow meter at zero flow conditions
The Journal of the Acoustical Society of America 142, 825 (2017); https://
doi.org/10.1121/1.4996851

AIP Advances 10, 105221 (2020); https://doi.org/10.1063/5.0022154 10, 105221

© 2020 Author(s).
 AIP Advances ARTICLE scitation.org/journal/adv

A comprehensive review on accuracy

in ultrasonic flow measurement using
reconfigurable systems and deep
learning approaches
Cite as: AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154
Submitted: 25 July 2020 • Accepted: 16 September 2020 •
Published Online: 13 October 2020

Senthil Kumar J,a) Kamaraj A,b) Kalyana Sundaram C,c) Shobana G,d) and Kirubakaran Ge)

AFFILIATIONS
Department of Electronics and Communication Engineering, Mepco Schlenk Engineering College, Sivakasi, Tamil Nadu, India

a)
[email protected]
b)
Author to whom correspondence should be addressed: [email protected]
c)
[email protected]
d)
[email protected]
e)
[email protected]

ABSTRACT
Flow rates of fuel are a major control variable in the engines of airborne vehicles. Accuracy in flow rate measurements became a mandatory
requirement for the testing and reliable operation of those engines. Flow rate measurement is also essential in food industries, automotive
industries, and chemical industries. Flow rate measurement using ultrasonic transducers is an appropriate choice because of its unique prop-
erties when they are in contact with the gases and water medium, as well as larger diameter pipelines. Estimating the echo signal of the
ultrasonic flowmeter is a challenging task; however, the processing tasks for real-time performance are mandatory for improving accuracy in
flow rate measurements. This article reviews the improvement in the accuracy of flow rate measurements of liquids and gases incorporating
the modern technological trends with the support of field programmable gate arrays, digital signal processors, other advanced processors, and
deep learning approaches. The review also elaborates on the reduction of uncertainty in single path and multi-path ultrasonic flowmeters.
Finally, future research prospects are put forward for developing low cost, reliable, and accurate ultrasonic flowmeters for extensive categories
of industrial applications.
© 2020 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license
(http://creativecommons.org/licenses/by/4.0/). https://doi.org/10.1063/5.0022154., s

I. INTRODUCTION Depending on the physical parameters to be measured, the flowme-

ters may be used for measuring liquid or gas by applying some
Generally, a flowmeter measures the amount of liquid, gas, or basic principles of physics and engineering to analyze the working
vapor moving through a pipe or conduit. The basic requirements for methodologies of flowmeters.
any flow measurement techniques are their abilities to calibrate and Studies have concluded the usage of digital meters over ana-
integrate flow fluctuation with pipes, improved accuracy, high turn- log positive feedback systems maintains appropriate flow tube oscil-
down ratio, low cost, low sensitivity to dirt particles, low-pressure lations.1 The development of flowmeters based on a field pro-
loss, no moving parts, and resistance to corrosion and erosion. There grammable gate array (FPGA) provides highly stable driving signals
are many types of flowmeters available for use in industrial automa- to probe the sensors and provides accuracy in flow estimations.2
tion. Most of them fall under the category of mechanical, pres- High-speed, highly reliable, and high-precision measurements of an
sure, thermal, optical, electromagnetic, and ultrasonic flowmeters. ultrasonic flowmeter are developed by Qi et al.3 FPGAs can also be

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-1

© Author(s) 2020
 AIP Advances ARTICLE scitation.org/journal/adv

configured as transit-time flowmeter time measurement circuits.4 ● The impact of an ultrasonic flowmeter on maximizing the
They are also used to collect the signal and process the acquired accuracy and performance in flow rate measurements is
data at high speed.5 The waveform of the excitation signals has to highlighted.
be stored in the read-only memory module of the FPGA chip. ● A comprehensive view of the usage of FPGAs and other
The usage of a digital signal processor (DSP) for the design processors and their significant contributions toward flow
and implementation of flowmeters increases the accuracy in the measurements are surveyed from the recent literature.
propagation time, which is obtained as the difference in measuring ● Accomplishments of flow rate measurements with the sup-
ultrasonic signals.6 DSPs are also being programmed and configured port of standard deep learning frameworks and deep neural
for reducing the measurement errors to obtain improved precision networks are discussed from the literature.
in calculating accurate and reliable flow measurements.7 The echo ● Future research directions are highlighted in the stream of
signal is sent into the DSP chip to be processed for obtaining the flow rate measurement with modern strategies and artificial
gas flow rate.5 A complete ultrasonic gas flowmeter needs ultra- intelligence.
sonic gas transducers, transmitting/receiving signal channel switch
circuits, driving signal generation circuits, sensor components, and C. Organization of this article
amplification circuits as vital components.
Deep learning architectures are inspired by the biological This review article is organized as follows: Sec. II discusses the
aspects of the senses similar to human brains. Computers and pow- evolution of conventional flowmeters and ultrasonic flowmeters in
erful graphical processing units (GPUs) can emulate such deep the market. It also elaborates on and summarizes the growth of
learning architectures. Human brains normally learn easier concepts ultrasonic flowmeters and the impact of ultrasonic technology on
first and hierarchically organize them to learn more sophisticated flow rate measurement applications of liquids and gases. Section III
and abstract ideas. Being inspired and driven by this learning tech- summarizes the utilization of the modern technology FPGA and
nique, investigators have devoted many efforts including multiple advanced processors for driving, supporting, and accurate compu-
levels of abstraction and processed them to solve complex compu- tation of flow rate measurements. Section IV discusses the learning
tational problems. Few researchers have utilized the power of deep approaches dealt with the aid of deep neural networks and their
learning architectures for flow rate measurements.8 With the volume impact on accuracy measurements in flowmeters. Section V elabo-
of data available and being trained, those deep learning architectures rates on the key findings and summary of this article toward accurate
are configured to classify and estimate the flow rates of liquids and flow rate measurements along with the future scope of research left
gases with high accuracy.9 as an open-end problem. Finally, Sec. VI concludes with the key
impact on the society. Figure 1 shows the overall organization of this
article.
A. The motivation of this survey and comparison
to other surveys
In this review, the intention is to provide an outline of the II. CONVENTIONAL FLOWMETERS AND EVOLUTION
progressive technologies in the ultrasonic based flowmeter measure- OF ULTRASONIC FLOWMETERS
ment techniques. Another significant aim is to summarize the major A. Magnetic flowmeters
emerging technologies such as the usage of FPGAs as reconfig-
urable systems and other advanced processors to support the accu- In magnetic flowmeters, the measured liquid must be water-
racy enhancement inflow measurements.10 Additional noteworthy based or conductive. This makes the magnetic flowmeter one of
discussions are focused on the usage of deep learning and deep the excellent choices for wastewater flow measurement. Magnetic
neural networks for the accurate flow rate estimation in ultrasonic meters are volumetric meters that have no moving parts. They are
flowmeters. ideal for those choices, where we need not be exposed to the mea-
On the other hand, a large number of applications were put sured liquid while working on the meter. The working logic of mag-
forward with the advancements in the technologies using modern netic flowmeters is based on Faraday’s law. Its main requirement
flowmeters. It is significant to consider the accuracy in the flow rate is the liquid under which the test is performed should be conduc-
of underlying modern technologies with analyses of the flow data tive. The measured voltage is dependent on the average velocity
observed from the systems. Table I summarizes the surveys related of the liquid, the strength of the magnetic field, and the length
to the usage of various flowmeters and their applications. The sum- of the conductor or the distance between the electrodes. Magnetic
mary reveals the significant contributions from those articles as well flowmeters are simplest in terms of their construction, and they
as their relevance to the proposed theme of the survey in this article. have no moving parts. However, they work only for conducive
measurements.
B. Contribution
B. Mechanical flowmeters
The major contributions of this review article are listed as
Mechanical flowmeters have some moving internal parts,
follows:
which may not fit with specific applications such as effluent waters
● The evolution of a flowmeter based on the different land- that carry larger particles. It may damage or obstruct the internal
scapes of flow measurements is studied. parts of the meters. The mechanical flowmeters have a rotational
● The key areas of flowmeter applications across different device inside such a paddle wheel or propeller. The liquid flowing
domains are identified from the state-of-the-art literature. through a pipe causes the rotation of the internal paddle, which

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-2

© Author(s) 2020
 AIP Advances ARTICLE scitation.org/journal/adv

TABLE I. Summary of surveys related to flowmeters and their applications.

Flowmeters/flow measurement Outcomes/relevance to reconfigurable

applications References Major contribution systems and deep learning approaches

A survey on the modern technolo- No explicit focus on the hardware used

11 gies in the process of tomography and the learning techniques. Only throm-
with multiphase flow measurement bus detection and cell sensing are done
A comprehensive survey of the soft Only trends in soft computing techniques
12 computing techniques for multi- are discussed. No explicit focus on the
phase flow metering FPGAs and the learning techniques
Multiphase flow measurement
A literature study for the discovery Only palm oil quality control phases are
of a suitable process for a palm oil explored. No significant contribution to
13
flow monitoring system the hardware used and the learning tech-
niques
A review of the investigation sta- Future trends of the downhole multi-
tus of the downhole multiphase flow phase flow measurement are analyzed. No
14
metering technology explicit focus on the hardware used and
the learning techniques

A survey on the clinical use of PET Physiological measurements of changes

to measure vascular parameters in in the tumor are explored. No significant
15
humans contribution to the hardware used and
the learning techniques

Blood flow measurement Studies on water PET and arterial No explicit focus on the hardware used
spin labeling to measure brain per- and the learning techniques. Accuracy
16
fusion of blood flow and reproducibility of arterial spin label-
ing are performed
Systematic literature on sponta- Focused on explorative analyses in blood
17 neous migraine attacks with blood flow velocity. No explicit application on
flow measurement the FPGAs and the learning techniques

A critical survey of the works avail- No significant contribution to the hard-

able using TTFM in CABG surgery ware used and the learning techniques.
18
Only limitations and varying threshold
values for TTFM are discussed
Transit-time flow measurement
Studies reported on abnormal grafts No explicit focus on the hardware used
or graft reviews on patients with and the learning techniques. The studies
19
heterogeneous data focus only on assessing diagnostic accu-
racy

This paper summarizes airflow No significant contribution to the hard-

measurement techniques including ware used and the learning techniques.
20
particle streak velocimetry Highlights only particle streak velocime-
try
This paper summarizes the tech- No explicit focus on the hardware used
Airflow measurement niques for the measurement of cool- and the learning techniques. With experi-
21
ing air volume flow in the vehicle mental values of CFD results, suggestions
are provided on airflow
A review on the overview of cur- No significant contribution to the hard-
rent endeavors to improve the air- ware used and the learning techniques.
22
cooling efficiency Thermal measurement issues alone are
discussed

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-3

© Author(s) 2020
 AIP Advances ARTICLE scitation.org/journal/adv

TABLE I. (Continued.)

Flowmeters/flow measurement Outcomes/relevance to reconfigurable

applications References Major contribution systems and deep learning approaches

This survey tends to explore No explicit focus on the hardware used

advanced measurement in this field and the learning techniques. The correla-
Two-phase gas/liquid flow measurement 23
of two-phase two-component flow tions of droplet profiles and bubbles are
investigated

A description of the calibration pro- Only the measurement of barometric and

cess of the capillary flowmeter mea- temperature issues is analyzed. No signif-
Capillary flowmeter 24
suring device icant contribution to the hardware used
and the learning techniques

A survey of the “bubble theory” that Examples are provided with combined
defines faults due to phase decou- phase decoupling and compressibility
25 pling of two-phase fluids errors. No significant contribution to the
hardware used and the learning tech-
Coriolis flowmeter
niques
This survey attempts to summarize No explicit focus on the hardware used
26 the available works on the Coriolis and the learning techniques. Only testing
mass flowmeter of CFM and flow dynamics are analyzed

A survey of the literature show- No explicit focus on the hardware used

ing consistency in the usage of a and the learning techniques. Free-flap
Doppler flowmeter 27
Doppler flowmeter monitoring is reported with the Doppler
flowmeter

A review on an eddy current Working with an eddy current flowmeter

flowmeter, under various test for different applications is analyzed. No
Eddy current flowmeter 28
conditions explicit focus on the hardware used and
the learning techniques

This paper highlights various Aspects of a vortex flowmeter develop-

aspects of the design of a vortex ment with its design are discussed. No
29
flowmeter significant contribution to the hardware
used and the learning techniques
Vortex flowmeter
In this paper, various aspects of No explicit focus on the hardware used
the problems concerning the vortex and the learning techniques. However,
30
meter are described the most powerful accomplishments in
this field are described

This paper surveys some important No significant contribution to the hard-

developments of transit-time ultra- ware used and the learning techniques.
Ultrasonic flowmeter 31
sonic flowmeters The custody and accurate measurement
in the flow of slurries are analyzed

creates a flow rate that is proportional to the rotational speed of difference between the applied pressures is measured before and
the internal paddle. The faster the paddle rotates, the more the flow after the primary element. A vibrating tube is present in the Cori-
occurring through the pipe, which in turn sends a graduated signal olis method, in which the fluid passes through the vibrating tube.
to the controller. The pipe size and temperature of the water flow- The amplitude of vibration will be proportional to the mass flow of
ing through the meter are correlated with the rotational speed with liquid. Using an open channel or variable area type of flow mea-
a unit measurement such as gallons per minute. surement meters ensures to calculate the flow rate from the veloc-
In the differential pressure type of flowmeters, the flow path ity, level, diameter, area, or depth information. All the categories
of liquid or gas is first blocked with a primary element and the of mechanical flowmeters have mechanical parts used as measuring

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-4

© Author(s) 2020
 AIP Advances ARTICLE scitation.org/journal/adv

FIG. 1. Overview of organization of this article.

FIG. 2. Categories of flowmeters based on the technology used for measurement.

elements, and however, the accuracy and lifetime are less due to the solve flow instabilities by considering the nonlinear dynamics of the
moving mechanical parts. fluids.

D. Ultrasonic flowmeters
C. Vortex flowmeters In this type of flowmeter, an ultrasonic signal is transmitted
Vortex flowmeters measure the vortices, which have a sensor downstream or in the direction of the flow, while another signal is
tab that will bend and flex from side to side as the vortex passes. transmitted upstream. The delta or the differential time is used to
The bend and flex action will then produce an output frequency calculate the velocity of the liquid in the medium. The estimated
proportional to the volumetric flow. In the vortex flowmeters, the velocity is then used to calculate the volumetric flow through the
amount of bending is proportional to the flow rate, and however, pipe.
the accuracy is less due to moving parts in vortex flowmeters, and it Ultrasonic flowmeters have numerous advantages compared to
also leads to a loss in signal flow during the measurement process. other categories of flowmeters. The usage of ultrasonic flowmeters
Particle image velocimetry (PIV) based flow measurement is used will not introduce any interference inflow of liquid. It is a good
in Taylor–Couette flow pattern experimentation in Ref. 32. It helps appropriate option for both gas and liquid measurements. However,

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-5

© Author(s) 2020
 AIP Advances ARTICLE scitation.org/journal/adv

Liquid and gas

the measurement can be improved by incorporating high-speed
processors.

Vortex

✓✓✓

✓✓
1:50
Hence, the ultrasonic flowmeter is a reliable meter for liquid

χ
χ
χ

χ
χ
and gas measurements. The summary of the types of flowmeters
based on the measurement is shown in Fig. 2. The characteristics
and performance of various categories of flowmeters are listed in
Table II. Based on Table II, magnetic and ultrasonic flowmeters are

Liquid and gas

having better performance characteristics. However, the mechan-

Coriolis

✓✓✓
✓✓✓

1:100
ical parts in the magnetic meter make it unsuitable for gas flow

✓✓
✓✓
✓✓

χ
χ
measurement.
The volumetric flow rate of liquids and gases is measured to
estimate the velocity of the fluid or gas in the medium. For record-
ing the rate of flow, various linear, non-linear, mass, and volu-

Liquid and gas

metric rates can be computed with appropriate flow measurement

Ultrasonic
devices. The flow rate F of any gas or liquid in any medium can be

✓✓✓
✓✓✓
✓✓✓

✓✓

✓✓
✓✓
1:20
expressed as

χ
√
2pc
F = Af Cf , (2.1)
d

Liquid and gas

where Af represents the flow area, Cf is the flow coefficient, d is

Variable area
the density of the fuel, and pc represents the control chamber pres-

✓✓✓
✓✓
1:10
sure. Based on the mentioned expression, flow rates are measured

χ
χ
χ

χ
χ
for various applications and methodologies listed in Table III.
Dynamic flowmeter calibration from small to medium flow
rates in aero-engine tests is done using computational fluid dynam-
ics (CFD) and a piezoelectric stack driven double nozzle–flapper
Turbine

✓✓✓
✓✓✓
Liquid
valve.33 The echo energy integral based signal processing method

1:20
χ
χ
χ

χ
χ
is proposed with calibration experiments performed to verify the
validity of maximum and minimum measurable flow rates using
the FPGA and DSP.34 An ultrasonic flowmeter was developed for

Liquid and gas

Differential

applications in cryogenic conditions; by using a steady-state Coriolis

pressure

✓✓✓
flowmeter, unsteady fluctuations of imposed disturbance are mea-

✓✓
1:10
χ
χ
χ

χ
χ
sured accurately.35 Flow measurements at high-temperatures up to
320 ○ C are performed using Z-mode and V-mode configurations in
a metallic wedge design of a pipe containing mineral oil.36
An ultrasonic gas flowmeter is developed with the DSP and
Conducting liquid Conducting liquid

FPGA for analyzing different flow rates using gas flow by generat-
(electrode)

ing echo signal envelopes in calibration experiments.37 A 2D flexural

Magnetic

✓✓✓
✓✓✓
✓✓✓
✓✓✓
✓✓✓

✓✓✓
1:100

✓✓
ultrasonic array transducer is developed to validate the flow mea-
surement using the spatial averaging method on a flexural array of
ultrasonic transducers.38 The flow rate of individual phases in gas–
liquid flows in a Venturi tube using a twin-plane capacitive sensor
TABLE II. Characteristics and performance of flowmeters.

is done to determine the liquid flow rate independent of the flow

patterns.79
(no electrode)

Fluid flow measurement with air bubbles was done as a single-

Magnetic

✓✓✓
✓✓✓
✓✓✓
✓✓✓
✓✓✓

✓✓✓
✓✓✓
1:20

phase flow measurement using an ultrasonic flowmeter at different

temperatures through PVC and steel pipes.39 Identification of two-
phase gas/liquid flow regimes using a Doppler ultrasonic sensor and
machine learning approaches was done for industrial practice.40,44
Horizontal oil–gas–water three-phase flow non-intrusive measure-
Obstruction in the pipe
Measurement principle

ment was done using ultrasound Doppler frequency shift in an oil–

Measurement objects

Effect of uneven flow

gas–water three-phase flow test loop system to obtain acceptable

Effect of adhesion

accuracy.41
Effect of solids

Coriolis two-phase flow metering solution for nitrogen/synthetic

Moving parts

Range ability
Pressure loss

oil mixtures errors is determined by comparing the reference single-

Accuracy

phase measurements based on a low dimensional polynomial of

the observed temperature suited to high viscosity oil/gas mix-
tures.42 The evaluation method for the ultrasonic flowmeter for

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-6

© Author(s) 2020
© Author(s) 2020
TABLE III. Summary of applications and methodologies used for flow measurement.
AIP Advances

S. no. References Application Simulation/analysis Experimentation Methodology Outcome

Aero-engine test Computational fluid Flow excitation process Piezoelectric stack Dynamic calibration
1 33 dynamics (CFD) experimentation driven double system for gear
nozzle–flapper valve flowmeters

Echo energy integral The ultrasonic echo Calibration experiments FPGA and DSP The maximum and
2 34 based signal processing energy gradient is are performed to verify minimum flow rates
method analyzed the validity improved

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154

Ultrasonic flowmeter Measuring flow rate Prototype was installed Unsteady The frequency and
for cryogenic conditions fluctuations in the in a flow test rig and state-particle image amplitude of unsteady
3 35
range of 0 Hz–70 Hz was velocimetry fluctuations are
validated measured

Flow measurements at Ultrasonic flowmeters Flowmeters in both Metallic wedge design Stable flow rate
4 36 high-temperatures transfer oil up to 320 ○ C Z-mode and V-mode measurement of
configurations mineral oil

Echo signal envelope Echo signal envelopes Gas flow calibration FPGA and DSP Quick and accurate
5 37 fitting under different flow experiments are location of the echo
rates are analyzed conducted signal

2D flexural ultrasonic Root-mean-square An array comprising The spatial averaging A wide range of
6 38 array transducer deviation of the average eight elements method is proposed industrial applications
flow velocity

The flow rate of Pressure data from a Combines a twin-plane Liquid Liquid flow rate,
7 79 individual phases in Venturi tube capacitive sensor rate-arithmetical independent of the
gas–liquid flows approach flow pattern
ARTICLE

Fluid flow measurement Single-phase flow Hydraulic bench used Transducers mounted Flow through the
with air bubbles measurement using to establish the fluid using the V-method PVC pipe was higher
8 39 ultrasonic flowmeter at flow loop in 20 mm and N-method than that through
diff. temperatures PVC and steel pipes the steel pipe at diff.
temperatures

Identification of Extract S-shaped riser, Machine learning Flow regime

9 40 two-phase gas/liquid frequency-domain connected to a approaches identification for
flow regimes features and classify horizontal pipe industrial practice
scitation.org/journal/adv

10, 105221-7
 AIP Advances ARTICLE scitation.org/journal/adv

Can be applied to
similar flow con-
ditions to obtain

Well suited to high

viscosity oil/gas mix-

The feasibility of the

gasotron is validated
the rectification effect of gasotron and its error indication in dif-

acceptable accuracy
ferent installation conditions of computational fluid dynamics are

for optimization
optimized.43
Outcome

III. FPGA AND ADVANCED PROCESSERS

FOR ULTRASONIC FLOWMETERS

tures
The ultrasonic sensor has a wide variety of applications. It not
only is limited to electronics but also has been used across all the
engineering fields. Table IV shows the general applications such as
observed temperature wind speed measurement and direction identification,53 irrigation
Computational fluid
Ultrasound Doppler

field for reducing the labor and water consumption,54 target detec-
calculated with the
Methodology

frequency shift in
three-phase flow

Fluid viscosity is

tion in military/navy and defense purposes,10 and distance mea-

surements with high accuracy for vehicle movement and accident
avoidance system.80
dynamics

Most commonly, the ultrasonic concept is used to measure the

flow rate of liquids in the pipes. It may be single-phase or double
phase flow.55 In all the liquid flow measurements, clamp-on types
of sensors are preferred for that purpose.56 The methods used for
measuring the flow rate fall under any one of the Doppler55 or time
difference49 categories. In both methods, either single or arrays of
installation conditions
loop, CWUD system
three-phase flow test
Experimentation

Indication errors in

sensors are used in the application. In addition, the performance

Two commercial

factors commonly taken into account are stability, accuracy, and

Coriolis meters,
Oil–gas–water

three different

speed of measurement. Some additional features are added as a value

addition for the flow measurement such as jitter and offset nul-
working

lifying57 web-based flow monitoring and leakage detection. Addi-

tionally, RAM,60 automatic gain amplifier (AGA), finite impulse
response (FIR), bus interfaces, arbitrary function generator (AFG),
and URT are incorporated for better processing.
The acquired data have to be processed to reach the design
Oil mass flow errors are
Superficial flow velocity

objective. From the review on the literature, it is observed that all of

Simulation/analysis

implementation steps
Rectification effect of

them are incorporated configurable devices such as FPGAs,49 micro-

using Doppler shift

controllers,54 and PLDs.58 This shows that all of them are inter-
gasotron and its

ested in having a dynamic configuration of the processing. On the

FPGA device, Xilinx47,48 and Altera45,52 products are used for faster
determined

processing and huge resources. The design work is carried out in

VHDL, Verilog, and C programming in terms of coding. However,
in certain advanced levels, soft processor cores such as MicroBlaze
Embedded Development Kit (EDK)49 and NIOS50 are used to adopt
more sophisticated algorithms such as the Kalman filter,49 FFT,80
split-spectrum processing,10 and spline algorithm.45
Coriolis two-phase flow

Evaluation method for

More interestingly, the National Instruments (NI PXIe-7965R)

ultrasonic flowmeter

FPGA is used to process 1.2 GB/s data from 225 arrays of sensors to
Application

metering solution
three-phase flow

measure the velocity of the liquid,46 in which the magnetohydrody-

Oil–gas–water

non-intrusive
measurement

namics (MHD) method is adopted. In addition, a bio-inspired visual

sensor (housefly) is developed as Micro-Air-Vehicles (MAVs) for
unmanned automatic vehicles in defense.47
In summary, an ultrasonic sensor is used as a multi-
purpose device in engineering solutions. The solutions are imple-
mented in programmable devices with added features. Some
special algorithms are incorporated to improve the performance
References

factors such as processing speed, accuracy, and flow rate detec-

TABLE III. (Continued.)

41

42

43

tion. Hence, the above literature analysis proves that the ultra-
sonic flowmeter is suitable for commercial applications, which is
adopted with programmable devices that would improve the per-
formance in real-time. Figure 3 lists various algorithms and tech-
S. no.

niques used along with FPGA devices for ultrasonic flow mea-
surement applications. Table IV shows the summary of FPGA
10

11

12

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-8

© Author(s) 2020
 AIP Advances ARTICLE scitation.org/journal/adv

TABLE IV. Summary of FPGA devices used for ultrasonic flow measurement applications.

References Type of FPGA Sensing hardware Technique Outcome

45 Altera Cyclone II Clamp-on meter Improved spline-based Reduces the running time
(EP2C8Q208) mounted on a algorithm effectually and has better
stainless-steel pipe stability

80 Altera’s Cyclone Ultrasonic Ultrasonic and FFT Reliability, debugging,

(EP1C3T14) receiver/transmitter modules real-time, and measuring
accuracy

46 NI PXIe-7965R National Nine linear arrays with Magnetohydrodynamics Enables a continuous and
Instruments 225 elements of piezo (MHD) and ultrasound near-real-time operation
transducers array Doppler of the measurement
velocimetry (UADV)

47 Virtex 2 (XC2V250) Elementary motion Bio-inspired visual sensor Fast dynamics of micro-
detector (EMD), (housefly) air-vehicles, artificial eye,
photoreceptors and avionic payload

48 Virtex 5 Ultrasonic piezoelectric Time interval High-resolution mea-

transducers measurement surement of flows with
Reynolds number from 6
to 20 000

49 MicroBlaze II processor Ultrasonic transducers Time difference method Improves the stability and
(32-bit RISC soft uses the Kalman filter to
processor) and reduce the interference of
®
Zynq -7000 SoC noise

50 Altera EP2C8T144C8 Ultrasonic flow sensor Doppler effect Accurate, fast, and stable
and automatic gain
amplifier

51 Altera IP Core Ultrasonic transceiver IP design Maximum clock fre-

quency, measurement
accuracy

60 Altera EP1K30TI144 Ultrasonic flowmeter Time difference method The sample speed is up to
sensor 80 MHz

52 Altera EP2C5T Ultrasonic flow metering Transit-time difference; High speed and precise
linear interpolation

devices from the literature used for ultrasonic flow measurement a sample deep learning model followed for the flow rate accuracy
applications. estimation using the flow rate parameters on the input side of the
deep neural network. Initial data would be raw data, which are sub-
ject to pre-processing, and they are shaped based on the input layer
IV. DEEP LEARNING FOR ACCURACY ESTIMATION configuration of the deep neural network model. Formatted data
IN ULTRASONIC FLOWMETERS flow through the neural network to estimate the accuracy of flow
Deep learning neural network models such as the convolu- parameters in the final output layer of the model.
tion neural network (CNN), long short-term memory (LSTM) and A deep learning approach and impedance measurements for a
autoencoders are used for estimating the features from the collected multichannel complex system are used for multiphase flow char-
data, to correlate with the flow of liquids and gases in the medium. acterization.8 Label-free analysis and evaluation are performed
Table V shows the summary of deep learning techniques used for using the deep learning-enabled holographic-reconstruction and
ultrasonic flow measurement applications along with the support- phase-recovery frameworks.59 For the network, saturation mod-
ing technologies and outcomes from the literature. Figure 4 shows els and electromagnetic data are used, to test the sensitivity

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-9

© Author(s) 2020
 AIP Advances ARTICLE scitation.org/journal/adv

FIG. 3. Algorithms and techniques with FPGA devices for ultrasonic flow measurement.

of the deep learning model to multiple electromagnetic compo- diagnostics of air-conditioning systems based on refrigerant flow
nents, generalization challenges, noises, and the ability to recon- using a variable flow refrigerant system.66
struct layers of a 3D CNN.9 Here, a U-Net deep learning net- A non-intrusive thermal flowmeter is developed to observe the
work is altered to suit the application, and it is trained and thermal distribution of the pipe and flow rates using machine learn-
validated. ing algorithms such as the K-nearest neighbor (KNN), DT, random
Deep learning algorithm based image analysis methods are forest, Ada, and gradient boost.76 The development of bubble detect-
used to classify the accelerated and stable trickle flow in the trickle ing algorithms is initiated by using the bubbly flow images from
bed while monitoring the flow transition.60 Droplet measurements the labeled bubble dataset to provide a strong benchmark of train-
are done by background extraction and value selection based on a ing data and algorithms for the development of advanced bubble
threshold using video processing software.74 Multiphase flow mea- detecting algorithms.67 Multichannel fluid nonlinear flow patterns
surements using soft computing techniques are reviewed along with are analyzed using a multivariate time series in a 50 mm diameter
individual phase fractions and phase flow rates.12 A seven-layer pipe to characterize flow patterns in the multichannel.77 An artificial
CNN is used for parameter measurement of gas–liquid two-phase neural network (ANN)-based teaching–learning-based optimization
flows in the small size pipe.12 Bubble generative adversarial net- technique measures the wellhead choke liquid flow rate using six
works and image processing algorithms are used for extracting flow different variables providing improved accuracy in prediction.68
parameters from the time-averaged distribution of bubbly flow for Recurrent R-U-Net and convolutional LSTM are trained to pre-
the network providing better improvement in the quality of estima- dict the accuracy in flow rates using channelized geological models
tion.61 The convolution neural network and machine learning are to predict dynamic subsurface flows.69 The liquid flow process is
used to develop flow adversarial networks for discriminating flow optimized using the flower pollination algorithm along with an ANN
rate and domain invariant features, and they provide better flow rate for testing of data subsets, and cross-validations are done to esti-
prediction.62,75 mate the accuracy and optimize the liquid flow process.70 Oil and gas
An ensemble Kalman filter is used for updates and for testing well tests are done using data processing procedures through data
the proposed deep LSTM approach, which shows that the predic- acquired from good sensors and the combination of the recurrent
tions provide better divergence and performance.63 A deep learn- neural network (RNN) with the CNN and LST network experiment
ing framework is developed with hidden fluid mechanics provid- on the data collected.71 Coriolis mass flowmeter vibration signal
ing a simplified methodology for 2D and 3D flow measurements to and mass flow rate samples are fed to the LSTM, RNN, and ANN
extract accurate flow information.64 Bubbly flow prediction is per- for evaluating the performance of a Coriolis mass flowmeter.72 Gas
formed on the high and low fidelity data for exploring the simula- valve leakage rate prediction is performed using a deep belief net-
tion errors using a deep feedforward neural network to estimate the work prediction model with the available plug valve, and ball valve
error in coarse mesh fluid dynamics for better industrial design flow signals are collected using acoustic emission technology. The pre-
prediction.65 A deep belief network is developed for improved fault diction model is proven to be superior in terms of accuracy in the

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-10

© Author(s) 2020
 TABLE V. Summary of deep learning techniques used for ultrasonic flow measurement applications.

Authors Application Simulation/analysis Experimentation Deep learning frameworks Outcome

© Author(s) 2020
8
Dang et al. Multiphase flow Impedance measurement The relation between Combines deep learning Complicated flow para-
AIP Advances

characterization spatial and diverse flow and impedance meters are measured
behaviors measurement

Göröcs et al.59 Imaging flow cytometer Ocean samples are Liquid pumped through a Deep learning-enabled Economic, better through-
measured microfluidic chip of holographic- put, and label-free analysis
0.8-mm-thickness reconstruction of water flow

Colombo et al.9 Electromagnetic Crosswell acquisition Sensitivity to U-Net deep learning Data-driven and physics-
monitoring setup electromagnetic network constrained reservoir

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154

components monitoring

Wang et al.60 On flow regime transition Accelerated and stable The transition of the flow Deep learning algorithm Liquid-rich and gas-rich
trickle flow regime in the trickle bed based image analysis boundaries are observed
method

Yan et al.12 Flow pattern identification By training the dataset Horizontal pipe with the A seven-layer CNN Measurement of gas–
inner diameter of 4.0 mm liquid two-phase flow

Fu61 Bubbly flow measurement 3D reconstruction Parameters from BGAN framework with A generic framework for
algorithm time-averaged distribution image processing bubbly flow image genera-
of bubbly flow tion

Hu et al.62 Gas–liquid multiphase Multiphase flow Flow adversarial networks CNN and machine Improved accuracy and
flows measurements learning better flow rate prediction

Loh et al.63 Natural gas wells Ensemble Kalman filter Two gas production wells Deep LSTM Predictions with better di-
are used for testing the vergence and performance
proposed approach
ARTICLE

Raissi et al.64 Flow visualization data Spatio-temporal domain Hidden fluid mechanics Deep learning framework Simplified methodology
assimilation for training using Navier–Stokes with hidden fluid for 2D and 3D flow
expressions mechanics measurements

Bao et al.65 Bubbly flow prediction High and low fidelity data Errors in fine mesh and Deep feedforward neural Flow prediction, physics
coarse mesh fluid network guided deep learning
dynamics methods

Guo et al.66 Refrigerant flow Fault diagnosis of AC Variable flow refrigerant Deep belief network Improved fault diagnostics
systems system with four faults and feature extractions

Fu and Liu67 Bubbly flow images Labeled bubble dataset for Aspect ratio, circularity, Bubble generative Benchmark of training
scitation.org/journal/adv

training rotation angle, and edge adversarial networks data and algorithms
ratio are monitored

10, 105221-11
© Author(s) 2020
AIP Advances

TABLE V. (Continued.)

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154

Authors Application Simulation/analysis Experimentation Deep learning frameworks Outcome
68
Choubineh et al. Wellhead choke liquid Temperature, oil and gas Three parameter model ANN A significant influence on
flow rate specific gravity are and six parameter model the critical flow rate
correlated

Tang et al.69 Subsurface flow problems Saturation maps and Randomized maximum R-U-Net and Predicted in channelized
dynamic pressure likelihood method convolutional LSTM geological models

Dutta and Kumar70 Liquid flow process Pipe diameter, liquid Flower pollination ANN Testing of data subsets
conductivity, and sensor algorithm and cross-validation is
voltages performed

Li et al.71 Oil and gas well tests Data processing on data RNN with CNN and LST RNN Economic good tests for
from good sensors network oil and gas industries

Zhang et al.72 Coriolis mass flowmeter Vibration signal and mass Signal processing is LSTM, RNN, and ANN The performance of the
flow rate performed on the Coriolis mass flowmeter is
collected data evaluated

Zhu et al.73 Valve leakage rate Prediction models for Plug valve and ball valve Deep belief network Prediction of gas leakage
ARTICLE

prediction three types of data signals rate in natural gas pipeline

scitation.org/journal/adv

10, 105221-12
 AIP Advances ARTICLE scitation.org/journal/adv

FIG. 4. A sample deep learning model for flowmeter accuracy measurement.

prediction of the gas leakage rate.73 2D CNNs are used to learn spa- REFERENCES
tial features and to capture temporal correlations from high-speed 1
M. Zamora and M. P. Henry, “An FPGA implementation of a digital Coriolis
images of combustion instability.78 mass flow metering drive system,” IEEE Trans. Ind. Electron. 55(7), 2820–2831
(2008).
2
A. Rehman, L. Sun, A. Latif, and A. Buzdar, “Distance and speed measurements
V. KEY FINDINGS AND SUMMARY OF THIS ARTICLE using FPGA and ASIC on a high data rate system,” Int. J. Adv. Comput. Sci. Appl.
In this paper, first, the evolution of conventional flowmeters 6(10), 273–282 (2015).
3
is summarized with their advantages and applications to identify N. Qi, Y.-Z. Song, S.-M. Fu, and X.-L. Jiang, “Circuit design of ultrasonic
flow sensor based on FPGA,” in Proceedings of the International Confer-
the ultrasonic flowmeter as a better choice for flow rate measure-
ence on Information Technology and Electrical Engineering 2018 (ICITEE 18),
ment. Next, the impact of the ultrasonic flowmeter on maximizing 2018.
the accuracy and performance in flow rate measurements is briefed. 4
Ł. Zaworski, M. Kowalski, and M. Zieliński, “Application of FPGA devices to
Following that, the usage of FPGAs/DSPs and their significant con- ultrasonic flowmeter,” Pomiary, Autom., Kontrola 57, 1511–1514 (2011).
tributions toward flow measurements are surveyed from the recent 5
W.-J. Zhu, K.-J. Xu, M. Fang, W. Wang, and Z.-W. Shen, “Mathematical model-
literature. In addition, the support of deep learning frameworks ing of ultrasonic gas flow meter based on experimental data in three steps,” IEEE
and deep neural networks for the flow rate measurements is dealt Trans. Instrum. Meas. 65(8), 1726–1738 (2016).
6
with along with their future research prospects for developing eco- S. Gerasimov, V. Glushnev, and M. Panov, “Applications of digital signal pro-
cessing methods in ultrasonic flowmeters,” in 2018 IV International Conference
nomic, reliable, and accurate flow measurable ultrasonic meters for
on Information Technologies in Engineering Education (Inforino) (IEEE, 2018),
widespread groups of manufacturing and commercial applications. pp. 1–6.
7
H. A. Christmann, D. R. Smith, B. L. Weaver, W. R. Betten, and R. A. Nazarian,
“Real-time DSP system for ultrasonic blood flow measurement,” in IEEE Interna-
VI. CONCLUSIONS tional Symposium on Circuits and Systems, New Orleans, LA, USA (IEEE, 1990),
Various industrial and commercial applications need flow mea- Vol. 3, pp. 2045–2048.
8
surement of liquid or gas. Based on their requirements, the mechan- W. Dang, Z. Gao, L. Hou, D. Lv, S. Qiu, and G. Chen, “A novel deep learning
ical, magnetic, pressure, optical, thermal, vortex, and ultrasonic framework for industrial multiphase flow characterization,” IEEE Trans. Ind. Inf.
15(11), 5954–5962 (2019).
meters are equipped for measurement. Since accuracy in flow mea- 9
D. Colombo, W. Li, E. Sandoval-Curiel, and G. W. McNeice, “Deep-learning
surement is the most significant factor in the present-day applica- electromagnetic monitoring coupled to fluid flow simulators,” Geophysics 85(4),
tions, mechanical, magnetic, and ultrasonic meters are performing WA1–WA12 (2020).
better. In terms of repeatability, ultrasonic meters are performing 10
J. Weber, E. Oruklu, and J. Saniie, “FPGA-based configurable frequency-diverse
much better as observed and summarized from the literature. The ultrasonic target-detection system,” IEEE Trans. Ind. Electron. 58(3), 871–879
performance of the ultrasonic meter is improved by incorporating (2011).
11
reconfigurable devices such as FPGAs, DSPs, and advanced proces- J. Yao and M. Takei, “Application of process tomography to multiphase flow
sors for performance measurement. The computation of accuracy in measurement in industrial and biomedical fields: A review,” IEEE Sens. J. 17(24),
8196–8205 (2017).
flow measurement can be much improved by adopting deep neu- 12
Y. Yan, L. Wang, T. Wang, X. Wang, Y. Hu, and Q. Duan, “Application of soft
ral networks in flow rate measurement. In the future, miniature computing techniques to multiphase flow measurement: A review,” Flow Meas.
bore ultrasonic sensors can be used for measuring the flow of liq- Instrum. 60, 30–43 (2018).
uid or gas using radar technology with high accuracy, stability, and 13
H. L. M. Ameran, E. J. Mohamad, S. Z. M. Muji, R. A. Rahim, J. Abdullah, and
repeatability for various applications. W. N. A. Rashid, “Multiphase flow velocity measurement of chemical processes

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-13

© Author(s) 2020
 AIP Advances ARTICLE scitation.org/journal/adv

36
using electrical tomography: A review,” in 2016 IEEE Conference on Systems, C. Wang et al., “Wedge design for high-temperature ultrasonic flow rate
Process and Control (ICSPC) (IEEE, 2019), pp. 130–135. measurement,” Sens. Actuators, A 298, 111585 (2019).
14 37
Z. Yongjian, M. Yongxin, Z. Tao, P. Yanzhi, and Z. Junlin, “Review of down- L.-B. Mu et al., “Echo signal envelope fitting based signal processing methods
hole multiphase flow measurement techniques,” Technol. Superv. Pet. Ind. (3), 12 for ultrasonic gas flow-meter,” ISA Trans. 89, 233–244 (2019).
(2016). 38
L. Kang et al., “Flow velocity measurement using a spatial averaging method
15
H. Anderson and P. Price, “Clinical measurement of blood flow in tumours with two-dimensional flexural ultrasonic array technology,” Sensors 19(21), 4786
using positron emission tomography: A review,” Nucl. Med. Commun. 23(2), (2019).
131–138 (2002). 39
N. Uwaezuoke et al., “Uncertainty in fluid flow measurement: A case study
16
A. P. Fan, H. Jahanian, S. J. Holdsworth, and G. Zaharchuk, “Comparison of of flow measurement comparison through PVC and steel pipes at varying tem-
cerebral blood flow measurement with [15O]-water positron emission tomogra- peratures for liquid with small air bubbles,” Adv. Pet. Explor. Dev. 17(1), 71–78
phy and arterial spin labeling magnetic resonance imaging: A systematic review,” (2019).
J. Cereb. Blood Flow Metab. 36(5), 842–861 (2016). 40
G. N. Somtochukwu et al., “Identification of gas-liquid flow regimes using a
17
N. A. Shayestagul, C. E. Christensen, F. M. Amin, S. Ashina, and M. Ashina, non-intrusive Doppler ultrasonic sensor and virtual flow regime maps,” Flow
“Measurement of blood flow velocity in the middle cerebral artery during sponta- Meas. Instrum. 68, 101568 (2019).
neous migraine attacks: A systematic review,” Headache: J. Head Face Pain 57(6), 41
C. Tan, X. Dong, and F. Dong, “Continuous wave ultrasonic Doppler modeling
852–861 (2017). for oil–gas–water three-phase flow velocity measurement,” IEEE Sens. J. 18(9),
18
L. Niclauss, “Techniques and standards in intraoperative graft verification by 3703–3713 (2018).
transit time flow measurement after coronary artery bypass graft surgery: A critical 42
M. Tombs, F. Zhou, and M. Henry, “Two-phase Coriolis mass flow metering
review,” Eur. J. Cardio-Thorac. Surg. 51(1), 26–33 (2017). with high viscosity oil,” Flow Meas. Instrum. 59, 23–27 (2018).
19 43
D. J. F. M. Thuijs, M. W. A. Bekker, D. P. Taggart, A. P. Kappetein, T. S. Peng et al., “Performance optimization of ultrasonic flow meter based on
M. Kieser, D. Wendt, G. Di Giammarco, G. D. Trachiotis, J. D. Puskas, and S. computational fluid dynamics,” Adv. Mech. Eng. 10(8), 1–9 (2018).
J. Head, “Improving coronary artery bypass grafting: A systematic review and 44
H. Shaban and S. Tavoularis, “Measurement of gas and liquid flow rates in two-
meta-analysis on the impact of adopting transit-time flow measurement,” Eur. J. phase pipe flows by the application of machine learning techniques to differential
Cardio-Thorac. Surg. 56(4), 654–663 (2019). pressure signals,” Int. J. Multiphase Flow 67, 106–117 (2014).
20
J. McWilliams, Review of Air Flow Measurement Techniques (Air Infiltration 45
L. Ma, J. Liu, and J. Wang, “Study of the accuracy of ultrasonic flowmeters for
and Ventilation Centre Operating Agent and Management INIV EEIG, 2002). liquid,” AASRI Procedia 3, 14–20 (2012).
21
T. Kuthada, “A review of some cooling air flow measurement techniques for 46
R. Nauber, N. Thieme, H. Beyer, L. Büttner, D. Räbiger, S. Eckert, and J. Czarske,
model scale, full scale and CFD,” SAE Int. J. Passenger Cars-Mech. Syst. 6, 88–96 “Modular ultrasound array Doppler velocimeter with FPGA-based signal pro-
(2013). cessing for real-time flow mapping in liquid metal,” Phys. Procedia 70, 537–540
22
J. Wan, X. Gui, S. Kasahara, Y. Zhang, and R. Zhang, “Air flow measurement (2015).
and management for improving cooling and energy efficiency in raised-floor data 47
F. Aubépart and N. Franceschini, “Bio-inspired optic flow sensors based on
centers: A survey,” IEEE Access 6, 48867–48901 (2018). FPGA: Application to micro-air-vehicles,” Microprocess. Microsyst. 31(6), 408–
23
M. A. Rahman, A. Amirfazli, T. Heidrick, and B. A. Fleck, “A review on 419 (2007).
advanced two-phase gas/liquid flow measurement techniques,” in Proceedings of 48
S. Grzelak, J. Czoków, M. Kowalski, and M. Zieliński, “Ultrasonic flow
the International Conference on Mechanical Engineering, 2007. measurement with high resolution,” Metrol. Meas. Syst. 21(2), 305–316
24
J. Topolnicki, M. Kudasik, N. Skoczylas, and J. Sobczyk, “Low cost capillary flow (2014).
meter,” Sens. Actuators, A 152(2), 146–150 (2009). 49
Y. Guiping and M. Xiaoming, “Economy value of high precision ultrasound
25
N. T. Basse, “A review of the theory of Coriolis flowmeter measurement errors flowmeter,” J. Phys.: Conf. Ser. 1237, 022009 (2019).
due to entrained particles,” Flow Meas. Instrum. 37, 107–118 (2014). 50
C.-H. Moon, “Implementation of an FPGA-based system for temperature, flow,
26
B. R. Binulal and J. Kochupillai, “Coriolis flow meter: A review from 1989 to and hydraulic pressure measurement of fluids,” Int. J. Mech. Eng. Rob. Res. 9(9),
2014,” Int. J. Sci. Eng. Res. 5(7), 718–723 (2014). 1288–1293 (2020).
27 51
J. C. Yuen and Z. Feng, “Monitoring free flaps using the laser Doppler B. J. Sun and K. Wang, “Ultrasonic flowmeter based on FPGA,” Appl. Mech.
flowmeter: Five-year experience,” Plast. Reconstr. Surg. 105(1), 55–61 Mater. 291-294, 2566–2569 (2013).
(2000). 52
Y. Bo, C. Li, and L. Yupin, “High-speed and precise measurement for ultra-
28
S. Poornapushpakala, C. Gomathy, J. I. Sylvia, B. Krishnakumar, and sonic liquid flow metering based on a single FPGA,” in IEEE Instrumentation and
P. Kalyanasundaram, “An analysis on eddy current flowmeter—A review,” in Measurement Technology Conference (IEEE, 2009).
Recent Advances in Space Technology Services and Climate Change 2010 (RSTS 53
K. Xie and K. Wang, “Measurement of wind speed and direction with ultrasonic
and CC-2010) (IEEE, 2010), pp. 185–188. sensor using FPGA,” Energy Procedia 12, 837–843 (2011).
29
A. Venugopal, A. Agrawal, and S. V. Prabhu, “Review on vortex flowmeter— 54
R. Sood, M. Kaur, and H. Lenka, “Design and development of automatic water
Designer perspective,” Sens. Actuators, A 170(1-2), 8–23 (2011). flowmeter,” Int. J. Comput. Sci., Eng. Appl. 3(3), 49–59 (2013).
30 55
G. L. Pankanin, “The vortex flowmeter: Various methods of investigating T. T. Nguyen, H. Murakawa, N. Tsuzuki, H. N. Duong, and H. Kikura, “Ultra-
phenomena,” Meas. Sci. Technol. 16(3), R1 (2005). sonic Doppler velocity profile measurement of single-and two-phase flows using
31
G. Rajita and N. Mandal, “Review on transit time ultrasonic flowmeter,” spike excitation,” Exp. Tech. 40(4), 1235–1248 (2016).
in 2016 2nd International Conference on Control, Instrumentation, Energy and 56
M. Orvatinia, “Improvement of accuracy in multi-path ultrasonic flow meters,”
Communication (CIEC) (IEEE, 2016), pp. 88–92. Sens. Transducers 231(3), 1–9 (2019).
32 57
R. J. Campero and R. D. Vigil, “Axial dispersion during low Reynolds number A. Hamouda, O. Manck, M. Hafiane, and N.-E. Bouguechal, “An enhanced tech-
Taylor-Couette flow: Intra-vortex mixing effects,” Chem. Eng. Sci. 52(19), 3303– nique for ultrasonic flow metering featuring very low jitter and offset,” Sensors
3310 (1997). 16(7), 1008 (2016).
33 58
B. Wang et al., “Simulation and experimental verification on dynamic calibra- P. Sundin, “Development of a computer interface for a clamp-on ultrasonic
tion of fuel gear flowmeters,” Measurement 138, 570–577 (2019). flow meter,” M.Sc. thesis, Embedded Electronics and Computer Systems, Tekniska
34
B. Liu, K.-J. Xu, L.-B. Mu, and L. Tian, “Echo energy integral based signal pro- Forskningsinstitut Utför Mätningar, 2007.
cessing method for ultrasonic gas flow meter,” Sens. Actuators, A 277(7), 181–189 59
Z. Göröcs, M. Tamamitsu, V. Bianco, P. Wolf, S. Roy, K. Shindo, and K. Yanny,
(2018). “A deep learning-enabled portable imaging flow cytometer for cost-effective, high-
35
V. Leontidis et al., “Experimental validation of an ultrasonic flowmeter for throughput, and label-free analysis of natural water samples,” Light: Sci. Appl.
unsteady flow,” Meas. Sci. Technol. 29(4), 045303 (2018). 7(1), 1–12 (2018).

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-14

© Author(s) 2020
 AIP Advances ARTICLE scitation.org/journal/adv

60 71
C. Wang, S. Li, Y. Yang, Z. Huang, J. Sun, J. Wang, Y. Yang, Z. Liao, and B. Jiang, Y. Li, R. Sun, and R. Horne, “Deep learning for well data history analysis,” in
“On flow regime transition in trickle bed: Development of a novel deep-learning- SPE Annual Technical Conference and Exhibition (Society of Petroleum Engineers,
assisted image analysis method,” AIChE J. 66(2), e16833 (2020). 2019).
61 72
Y. Fu, “Development of advanced image processing algorithms for bubbly flow Y. Zhang, Y. Liu, Z. Liu, and W. Liang, “Developing a long short-term memory-
measurement,” Ph.D. dissertation (Virginia Tech, 2018). based signal processing method for Coriolis mass flowmeter,” Measurement 148,
62 106896 (2019).
D. Hu, J. Li, Y. Liu, and Y. Li, “Flow adversarial networks: Flowrate prediction
73
for gas-liquid multiphase flows across different domains,” in IEEE Transactions S.-B. Zhu, Z.-L. Li, S.-M. Zhang, and H.-F. Zhang, “Deep belief network-based
on Neural Networks and Learning Systems (IEEE, 2019). internal valve leakage rate prediction approach,” Measurement 133, 182–192
63 (2019).
K. Loh, P. S. Omrani, and R. van der Linden, “Deep learning and data assimi-
74
lation for real-time production prediction in natural gas wells,” arXiv:1802.05141 P. Hadikhani, N. Borhani, S. Mohammad, H. Hashemi, and D. Psaltis, “Learning
(2018). from droplet flows in microfluidic channels using deep neural networks,” Sci. Rep.
64 9(1), 1–7 (2019).
M. Raissi, A. Yazdani, and G. E. Karniadakis, “Hidden fluid mechanics: A
75
Navier-Stokes informed deep learning framework for assimilating flow visualiza- Z. Yang, H. Ji, Z. Huang, B. Wang, and H. Li, “Application of convolu-
tion data,” arXiv:1808.04327 (2018). tion neural network to flow pattern identification of gas-liquid two-phase flow
65 in small-size pipe,” in 2017 Chinese Automation Congress (CAC) (IEEE, 2017),
H. Bao, J. Feng, D. Nam, and H. Zhang, “Computationally efficient CFD pre-
diction of bubbly flow using physics-guided deep learning,” arXiv:1910.08037 pp. 1389–1393.
76
(2019). J. Amaral, J. R. Castro Silva, D. S. M. de Andrade, L. T. Ferreira, T. Motta
66
Y. Guo, Z. Tan, H. Chen, G. Li, J. Wang, R. Huang, J. Liu, and T. Ahmad, “Deep Quirino, and J. M. Quirino, “Machine learning algorithms applied to the infer-
learning-based fault diagnosis of variable refrigerant flow air-conditioning system ence of the flow rate in a non-intrusive thermal flowmeter,” in 2019 4th Interna-
for building energy saving,” Appl. Energy 225, 732–745 (2018). tional Symposium on Instrumentation Systems, Circuits and Transducers (INSCIT)
67 (IEEE, 2019), pp. 1–6.
Y. Fu and Y. Liu, “BubGAN: Bubble generative adversarial networks for synthe-
77
sizing realistic bubbly flow images,” Chem. Eng. Sci. 204, 35–47 (2019). Z. Gao, X. Hong, W. Dang, L. Hou, and M. Liu, “Multiresolution multiplex
68 network for analyzing multichannel fluid flow signals,” in IEEE Transactions on
A. Choubineh, H. Ghorbani, D. A. Wood, S. Robab Moosavi, E. Khalafi, and
E. Sadatshojaei, “Improved predictions of wellhead choke liquid critical-flow rates: Circuits and Systems II: Express Briefs (IEEE, 2019).
78
Modelling based on hybrid neural network training learning based optimization,” T. Gangopadhyay, A. Locurto, J. B. Michael, and S. Sarkar, “Deep learning algo-
Fuel 207, 547–560 (2017). rithms for detecting combustion instabilities,” in Dynamics and Control of Energy
69 Systems (Springer, Singapore, 2020), pp. 283–300.
M. Tang, Y. Liu, and L. J. Durlofsky, “A deep-learning-based surrogate model
79
for data assimilation in dynamic subsurface flow problems,” arXiv:1908.05823 A. D. N. Wrasse, D. Bertoldi, E. N. Dos Santos, R. E. Morales, and M. J. Da
(2019). Silva, “Gas liquid flow rate measurement using a twin-plane capacitive sensor and
70 a Venturi meter,” IEEE Access 7, 135933–135941 (2019).
P. Dutta and A. Kumar, “Modeling and optimization of a liquid flow process
80
using an artificial neural network-based flower pollination algorithm,” J. Intell. M. Wu and J. Yang, “APOFDM system design based on all phase FFT technol-
Syst. 29(1), 787–798 (2018). ogy,” J. Electron. Inf. Sci. 2(2), 72–75 (2017).

AIP Advances 10, 105221 (2020); doi: 10.1063/5.0022154 10, 105221-15

© Author(s) 2020