Radiation Physics and Chemistry 223 (2024) 112018

Contents lists available at ScienceDirect

Radiation Physics and Chemistry

journal homepage: www.elsevier.com/locate/radphyschem

Calcium carbonate scale thickness prediction in annular three-phase flow

using gamma-ray densitometry and artificial neural networks
A.C.L. Carvalho a , W.L. Salgado b , R.S.de F. Dam b, c , C.C. Conti d , C.M. Salgado b, *
a
Angra 2 Operation Department - DOD.O, Eletronuclear S.A., Rodovia Procurador Haroldo Fernandes Duarte, BR 101/RJ, km 521, 23948-000, Itaorna, Angra dos
Reis, Brazil
b
Division of Radiopharmaceuticals - DIRAD, Institute of Nuclear Engineering - IEN, Rua Hélio de Almeida 75, 21941-906, Cidade Universitária, RJ, Brazil
c
Nuclear Engineering Program - PEN/COPPE, Federal University of Rio de Janeiro - UFRJ, Avenida Horácio de Macedo 2030, G – 206, 21941-914, Cidade
Universitária, RJ, Brazil
d
Laboratory of Nuclear Instrumentation - LIN/COPPE, Federal University of Rio de Janeiro - UFRJ, Centro de Tecnologia (CT), Bloco I, Sala I-133, 21941-972, Cidade
Universitária, RJ, Brazil

A R T I C L E I N F O A B S T R A C T

Handling Editor: Dr. Chris Chantler This study presents a methodology based on gamma-ray densitometry for predicting calcium carbonate (CaCO3)
scale thickness in pipelines of three-phase systems (oil, salty water, and gas) in the petroleum industry. Both
Keywords: concentric and eccentric scale thicknesses were calculated using analytical equations and artificial neural net­
Calcium carbonate scale works. The artificial neural networks were trained using a backpropagation algorithm with data acquired from
Artificial neural network
the Monte Carlo N-Particle (MCNP6) computer code. The model utilized to achieve the supervised training of the
MCNP6 code
network consisted of three 1¼ × ¾″ NaI(Tl) detectors positioned 120◦ apart around the pipe. Results demonstrate
Multiphase flow
that the networks successfully predicted over 96 % of both concentric and eccentric cases of CaCO3 scale
thickness with a relative error within 10%.

1. Introduction removing dissolved carbon dioxide (CO2) (Bertran et al., 2002). CO2
increases the solubility of CaCO3 by forming a more soluble bicarbonate
During petroleum extraction, a three-phase flow consisting of oil, (Ca(HCO3)2) through the reversible reaction: CaCO3 + CO2 + H2O ⇆ Ca
seawater, and gas is brought to the surface due to its generation process, (HCO3)2. Adverse changes in the conditions of the oil extraction process
which involves the diagenesis, catagenesis, metagenesis and meta­ may favor the release of CO2, thus shifting the reaction equilibrium to
morphism phases, as well as the migration of water from aquifers to the the left, leading to the precipitation of CaCO3 (Lopes et al., 2003). In
reservoir rock (Tissot and Wellte, 1984; Triggia et al., 2004). The other words, during petroleum extraction, reduced pressure and
extraction process is marked by the emergence of scales: insoluble pre­ increased temperature move CO2 from the oil and gas phases to the
cipitates caused by changes in thermodynamics, chemical and micro­ water. This alters the equilibrium of the reaction, decreasing carbon
biological reactions, and water evaporation (Kamal et al., 2018). Scales solubility in the solution precipitating CaCO3.
are the major operation problem in the petroleum industry, directly Several studies on CaCO3 scale have been conducted over the years,
related to corrosion equipment, flow reduction, and, in some cases, including the prediction of the deposition tendency (Stiff and Davis,
production interruption (Olajire, 2015; Kamal et al., 2018). The most 1952; Sippel and Glover, 1964; Lindlof and Stoffer, 1983; Kumar et al.,
frequently encountered types of scales in the offshore industry are the 2018), correlation of flow conditions with other parameters like tem­
sulfides, oxides, sulfates, and carbonate scale (Kamal et al., 2018). perature, pressure, and velocity (Diebes and Jessen, 1967; Watkinson
Calcium carbonate scales are the most common type of carbonate scale and Martinez, 1975; Mitchell et al., 1980; Allaga et al., 1992; Moghadasi
and can appear as calcite, vaterite or aragonite. The formation of cal­ et al., 2002), and analysis of the damage caused by the scales (Read and
cium carbonate (CaCO3) deposits occurs due to the depressurization of Rigen, 1982). Most of the mathematical models used to determine scale
water present in oil reservoirs, which alters the HCO−3 /CO−3 2 balance by thickness in pipes are based on thermodynamics and kinetics, requiring

* Corresponding author.
E-mail addresses: [email protected] (A.C.L. Carvalho), [email protected] (W.L. Salgado), [email protected] (R.S.deF. Dam), conti.
[email protected] (C.C. Conti), [email protected] (C.M. Salgado).

https://doi.org/10.1016/j.radphyschem.2024.112018
Received 24 October 2023; Received in revised form 17 June 2024; Accepted 30 June 2024
Available online 1 July 2024
0969-806X/© 2024 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
A.C.L. Carvalho et al. Radiation Physics and Chemistry 223 (2024) 112018

Fig. 1. Scale thickness defined by the segment AB: a) concentric; b) eccentric.

extensive knowledge of the flow itself, including temperature, pressure, successfully used to map linear and non-linear functions (Sekoguchi
pH, and ionic concentration. However, these parameters differ across oil et al., 1987; Cai et al., 1994; Mi et al., 1996; Salgado et al., 2010; 2020,
fields and may vary during the oil extraction and refinement process 2021a). Therefore, the MLP networks can be used to interpret the pulse
(Olajire, 2015). Therefore, radiation techniques have been applied in the height distribution (PHD) recorded with the gamma radiation detectors
offshore industry to overcome these obstacles. The main techniques to predict scale thickness. Nevertheless, the ANN must be supplied with
include computed radiography (Candeias et al., 2014), radiotracers a data set for the learning phase. For this purpose, Monte Carlo N-Par­
(IAEA - International Atomic Energy Agency, 2013), gamma densitom­ ticle 6 (MCNP6) is an important tool that can be utilized (Goorley et al.,
etry (Oliveira et al., 2015; Salgado et al., 2020; 2021a; IAEA - Interna­ 2016). MCNP6 is a computer code to simulate the radiation transport
tional Atomic Energy Agency, 2013) and neutron scanning (IAEA - process and the interaction of radiation within different materials. This
International Atomic Energy Agency, 2013). Gamma densitometry is code is widely used in various areas, such as nuclear reactors, dosimetry,
based on measuring the transmission of a gamma-ray beam to determine radiation detectors projects, nuclear medicine, particle accelerators, and
material density. It utilizes at least one detector to register counts from a radiation protection. The MCNP6 code for gamma-ray simulation in­
gamma-ray source passing through a pipe containing fluids. This tech­ cludes the photoelectric effect, Compton and Rayleigh scattering, Pair
nique has many applications beyond density prediction (Achmad and production, X-ray fluorescence, and bremsstrahlung (Goorley et al.,
Hussein, 2004; Roshani et al., 2013; Salgado et al., 2016), such as flow 2016).
measurement (Bishop and James, 1992; Hussein and Han,1995; Roshani This study presents a model for predicting the thickness of both
et al., 2017, 2018; Salgado et al., 2009, 2010; Affonso et al., 2020; Islami concentric and eccentric CaCO3 scales in the offshore industry, regard­
Rad et al., 2022), detection of the interface region, and identification of less of fluid presence. In the case of eccentric scale, the prediction is
fluids transported by polyducts (Salgado et al., 2021b), flow rate using made independently of its location within the pipe. Additionally, this
radioactive particle tracking (Dam et al., 2021), gamma-ray scanning in paper introduces a set of analytical equations for theoretically validating
pipes to locate blockages, scale or core buildup, corrosion, liquid in the mathematical simulation, a novel contribution to the field. To ensure
vapor lines, areas of lost refractory or lining in a pipe, and evaluation of applicability to real-world scenarios in the offshore industry, a simula­
catalyst fluidization (IAEA - International Atomic Energy Agency, 2020). tion of annular three-phase flow regime (oil, salty water, and methane
Gamma densitometry offers further advantages as it is a non-invasive gas) in a pipe with NaI(Tl) detectors was conducted using the MCNP6
technique. The detector installed outside the pipeline does not have code. The calculated PHDs for thickness served as patterns for training,
contact with the fluids inside, reducing costs and allowing for online testing, and validating the ANNs. Furthermore, the models developed in
monitoring. Therefore, gamma-ray densitometry is a promising tech­ the MCNP6 code for both concentric and eccentric scales were validated
nique for measuring scale thickness (Oliveira et al., 2015; Salgado et al., using analytical equations.
2020, 2021a; IAEA - International Atomic Energy Agency, 2013).
Nevertheless, the analytical equations used to determine scale 2. Methodology
thickness rely on specific measurement geometries, types of fluids in the
flow, and flow regime, which can change during the petroleum extrac­ In this study, the Monte Carlo N-particle (MCNP6) code was used to
tion and refining. Furthermore, scale thickness prediction is a complex simulate a gamma scattering system to calculate the calcium carbonate
process, particularly in cases with eccentric formation, due to non- (CaCO3) scale thickness through analytical equations and ANNs for both
uniform deposition (Salgado et al., 2021a). An alternative to over­ concentric and eccentric deposition. The mass attenuation coefficients
coming these obstacles is the use of artificial neural networks (ANNs), (μ/ρ) were obtained from the National Institute of Standards and
which are learning machines that simulate the learning mechanism in Technology (Hubbell and Seltzer, 1997). The dataset for the ANNs were
biological organisms (Aggarwal, 2018) and consist of simple processing obtained using the MCNP6 computer code.
units with the ability to store knowledge obtained experimentally
(Haykin, 2005). The multilayer perceptron (MLP) networks have been

2
A.C.L. Carvalho et al. Radiation Physics and Chemistry 223 (2024) 112018

Cp – radiation path through the pipeline (cm).

The scale thickness was determinate by the AB segment, as illus­
trated in Fig. 1, and calculated using Equation (5). The coordinates of its
endpoints were calculated using Equation (6), 7, 8, and 9.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
AB = (xa − xb )2 + (ya − yb )2 Equation 5

CD − AD
xa = Equation 6
2

ya = h Equation 7
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
Fig. 2. Trigonometric relations to calculate the radiation path in: a) scale plus xb = r 2 − yb 2 Equation 8
fluid and b) pipe.
yb = mxb Equation 9
2.1. Calculation of the CaCO3 scale thickness Where:
m – slope of line OB.
The calculation of the CaCO3 scale thickness is based on gamma The aim of this step is to validate the analytical equations consid­
transmission technique, where the system comprising the pipeline, ering the scale at 0◦ . In this configuration, the segments AB and AD
scale, and fluid is considered as an n-layers absorber. In this context, the shown in Fig. 1 represent the same value (see Fig. 1b); that is, the ra­
attenuation of the incident beam (I0 ) will vary for each material inside diation path in the AD segment is the maximum eccentric scale thickness
the pipe. Thus, the Beer-Lambert equation can be expressed as Equation that can be calculated using Equation (4).
(1). The study addressed the uncertainty in the scale thickness by
employing the uncertainty propagation law applied to Equation (10),
I = I0 . e− [(μp Cp )+(μf Cf )+(μs Cs )] Equation 1
after mathematical manipulations of Equation (5), and considering h =
Where: 0. The goal was to estimate the total uncertainty in the concentric scale
I – intensity of the emerging beam (γ.cm− 2. s− 1); thickness measurement. To accomplish this, partial derivatives corre­
I0 – intensity of the incident beam (γ.cm− 2. s− 1); sponding to the uncertainty of each variable involved in the calculations
μ – linear attenuation coefficient (cm− 1); were utilized, as specified by the Joint Committee for Guides in
Cn – radiation path to the material n (cm); Metrology (JCGM - Joint Committee for Guides in Metrology, 2008).
p, f, s – pipe, fluid, and scale. ⎛ ⎞
ln (k) − μp (2R − 2r) + μf 2r
The pipeline scanning, aimed at recording the relative counts of AB = ⎝ ( ) ⎠ Equation 10
gamma radiation through the pipeline, is conducted using a detector and 2 μf − μs
a source positioned diametrically opposite. Gamma ray attenuation
varies with each source-detector position, enabling the acquisition the of The partial derivatives for each variable involved in the calculations
the pipe’s content profile. These outcomes are utilized to determinate were calculated using Equations (11)–(15), and 16, with each equation
the scale thickness, denoted by the AB segment in Fig. 1 for concentric corresponding to a specific variable contributing to the determination of
and eccentric cases. It is important to note that the scanning procedure the total uncertainty in concentric scale thickness.
described in Section 2.1 was not carried out for eccentric scales. ∂AB r − R
Where: = Equation 11
∂μp μf − μs
R – pipe outer radius;
r – pipe inner radius;
∂AB ln(k) − μp (2R − 2r) + μf 2r − 1
h – step height for positions of source-detection system; = ⋅( )2 Equation 12
∂μs 2
CD – radiation path through the scale plus fluid; μ f − μs
AD – radiation path through the scale;
⎧ ( ) [ ] ⎫
AB – scale thickness. ⎪
⎨2r⋅2 μ f − μ s − ln(k) − μ p (2R − 2r) + μ f 2r ⋅2⎪
⎬
The radiation path through the scale + fluid and pipe, to each po­ ∂AB
= ( )2 Equation 13
sition of the source-detector system as depicted in Fig. 1a, can be ∂μf ⎪ ⎩ 4 μ f − μs
⎪
⎭
calculate using trigonometric relations, as illustrated in Fig. 2, and
described by Equation (2) and Equation (3) respectively. These equations ∂AB 1 1
were employed to determinate the radiation path through the scale, as =( )⋅ Equation 14
demonstrated by Equation (4).
∂k μf − μ s k
[√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅]
( ) − μp
CD = 2 r2 − h2 Equation 2 ∂AB
= Equation 15
∂ R μf − μ s
[(√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅) (√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅)]
Cp = 2 R2 − h2 − r2 − h2 Equation 3 ∂AB μp + μf
= Equation 16
∂r μf − μs
ln(k) + μp Cp + μf CD
AD = Equation 4 After applying the error propagation law, Equation (17) was derived
− μs + μf to describe the total uncertainty associated with concentric scales.
Where:
k – I/I0 ;

3
A.C.L. Carvalho et al. Radiation Physics and Chemistry 223 (2024) 112018

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) ( ) ( ) ( ) ( ) ( )̅
∂AB 2 ∂AB 2 ∂AB 2 ∂AB 2 ∂AB 2 ∂AB 2
σ AB = σk 2
+ σ μP 2 + σR 2
+ σr 2 + σ μS 2 + μμF Equation 17
∂K ∂μP ∂R ∂r ∂μS ∂μF

2.2. Detection geometry to calculate the scale thickness

The mathematical model developed for the simulations using the

MCNP6 code consisted of an AISI 316L stainless steel pipe with an
external diameter of 252 mm and a wall thickness of 2 mm. It initially
had a 5 mm thickness for the CaCO3 concentric scale, and the simulated
fluids inside the pipe were methane, oil, and salty water.
Fig. 3 illustrates the setup used to simulate the scanning procedure,
considering its symmetry, from h = 0 mm to h = 125 mm, gradually
increased by 5 mm steps. To obtain the current integrated over the
surface detector, the F1 tally card available with the MCNP6 code was
applied on a circular surface (detector) with a radius of 15 mm dia­
metrically opposite a pencil-beam Cs-137 point source.
To ascertain the smallest thickness applicable in analytical equa­
tions, initially, 1 mm of CaCO3 was considered, gradually incremented
in steps of 1 mm. For this purpose, only the height h = 0 mm was
considered at this stage.
The second part of the study consisted of evaluating three concentric
Fig. 3. Detection geometry for determining the radiation path through scale thickness, which were 5 mm, 48 mm, and 96 mm. Pipe scanning
the scale. was conducted for the detection heights ranging from h = 0 mm to h =

Fig. 4. Maximum eccentric scale thickness positioned at: a) 0◦ , b) 90◦ , c) 120◦ .

4
A.C.L. Carvalho et al. Radiation Physics and Chemistry 223 (2024) 112018

Table 1 but with maximum scale thickness of 5 mm, 50 mm, and 103 mm. This
Characteristics of the materials and fluids used in the MCNP6 code simulations. study investigated low, intermediate, and high values of scale used in
Weight fraction of the materials and fluids Density (g.cm− 3) μ (cm− 2. g− 1)a neural network training.
Steel Stainless 316L Mexican Crude Oil Salty water
8.0 0.975 1.04 2.3. Materials
7.358⋅10− 2a 8.507⋅10− 2a 8.439⋅10− 2a

C: 0.00030 H: 0.104039 H: 0.107423 The input file (INP) containing a model of the measurement geom­
Si: 0.01000 C: 0.853733 O: 0.853577 etry data to be used in the MCNP6 code requires information about the
P: 0.00045 S: 0.042228 Na: 0.1015735
S: 0.00030 Cl: 0.024265
chemical composition and density from the materials used for the sim­
Cr: 0.17000 ulations. Table 1 presents the weight fraction and density of all materials
Mn: 0.02000 and fluids employed in the simulations with the MCNP6 code (McConn
Fe: 0.65395 et al., 2011).
Ni: 0.12000
Mo: 0.02500

CaCO3 Air CH4 2.4. Compton effect

3 4
2.8 1.205⋅10− 6.67⋅10−
The Compton effect takes place when the incident gamma-ray in­
2a 2a
7.749⋅10− 7.713⋅10−

C: 0.120003 C: 0.000124 C: 0.25000 teracts with an electron in the absorber material (pipe, fluid, and scale),
O: 0.479554 N: 0.755268 H: 0.75000 transferring a part of its energy to this electron. Consequently, the
Ca: 0.400443 O: 0.231781 electron undergoes a deflection by an angle θ from its original direction.
Ar: 0.012827
The energy transferred to the electron and the scattering angle can be
NaI(Tl) MgO Al related through Equation (18). Moreover, the transferred energy can
3.667 3.58 2.7
range from zero to a significant fraction of the gamma-ray energy (Knoll,
Na: 0.8449 Mg: 0.603036 Al: 1.0 1999).
I: 0.1531 O: 0.396964
Tl: 0.0020 hv
Eγ = Equation 18
a
Mass attenuation coefficient used to analyze the materials/fluids in this 1 + mhv
0c
2 (1 − cos θ)

study.
Where:
hv – incident photon energy (eV);
125 mm, incremented in steps of 5 mm. In all these cases, the simulated m0 c2 – electron’s rest energy (eV);
fluids inside the pipeline were methane, oil, and salty water. θ – scattering angle.
To determine the maximum scale thickness for the eccentric cases, Compton scattering plays a significant role in this study because
the radiation path was determined solely for height h = 0 mm, with the relying solely on the transmitted gamma flux would be impractical. This
maximum eccentric scale thickness positioned at 0◦ , 90◦ , and 120◦ , as is primarily due to the unknown location of the maximum deposition of
illustrated in Fig. 4a and b, and 4c respectively. The fluids inside the eccentric CaCO3 scale, stemming from the non-uniformity of the depo­
pipeline were the same as those used for simulations of concentric cases, sition (Salgado et al., 2021a). Furthermore, Compton scattering is the

Fig. 5. Scale measurement geometry: a) concentric; b) eccentric.

5
A.C.L. Carvalho et al. Radiation Physics and Chemistry 223 (2024) 112018

Fig. 7. ANN architecture proposed for scale prediction.

previous study, where its dimensions were determined using the gam­
magraphy technique (Salgado et al., 2012, 2022).
The simulations were divided into two parts. The first part, con­
cerning deposition of the concentric scale, comprised 37 simulations.
The scale thickness varied from 2 mm to 124 mm, gradually increasing
by steps of 2 mm for the first 12 simulations and 4 mm for the remaining
25 simulations.
The second part focused on deposition of the eccentric scale. The
Fig. 6. NaI(Tl) detector model used for MCNP6 code simulations. CaCO3 scales were arranged inside the pipeline 15◦ intervals, resulting
in 24 locations for the maximum scale thickness. Forty simulations were
most prominent interaction mechanism for photons in the energy range conducted for each location, with the scale thickness ranging from 2 mm
of 100 keV to 10 MeV and for materials with an atomic number below to 164 mm. In these cases, the thickness was gradually increased by step
40. The probability of Compton scattering is directly proportional to the of 2 mm for first 13 simulations and 4 mm for the remaining 27 simu­
photon energy and inversely proportional to the atomic number of the lations. The training data set for the ANN on the eccentric cases
material with which the photons interact. The intensity of Compton comprised 960 simulations.
scattering can be calculated (Klein and Nishina, 1929). In this study, the It is noteworthy that, in the eccentric simulations, methane gas was
MCNP6 code was used to estimate the scattered events. The counting rate of used as the gaseous phase to better simulate the real conditions. Fig. 5
Compton scattering events depends on the density of the material under illustrates the setup used to simulate the concentric and eccentric cases.
analysis, as the Klein-Nishina differential cross section remains constant Fig. 6 provides further details about the NaI(Tl) detector model used
for a fixed geometry, incident photon energy, and photon flux intensity. for the MCNP6 code simulations. It comprises a homogeneous cylinder
crystal of sodium iodine doped with thallium, measuring 31 mm in
diameter and 19 mm in thickness. Additionally, the detector includes a
2.5. Detection geometry for the ANN training data reflective layer of magnesium oxide (MgO) and a housing of aluminum
(Al). These dimensions were determined through gammagraphy (Sal­
The mathematical model developed for the MCNP6 simulations gado et al., 2012).
comprised an AISI316L stainless steel pipe with an outer diameter of The adjustment of the Pulse Height Distribution (PHD) to a Gaussian
252 mm and a wall thickness of 2 mm. Inside the pipe, an annular three- response curve was carried out using the Gaussian Energy Broadening
phase flow regime was simulated, consisting of 60% Mexican crude oil (GEB) command, which is available in the MCNP6 code. The GEB pa­
(McConn et al., 2011), 30% salty water (4% NaCl), and 10% methane rameters were determined based on the full width at half maximum
gas. In this study, the volumetric fractions were held constant. However, (FWHM) versus energy curve, derived from experimental measurements
the effects of different volumetric fractions of fluids were explored in a of radioactive sources (Salgado et al., 2012).
previous article (Dam et al., 2022). Three 1¼ × ¾″ NaI(Tl) scintillation In both simulation cases (concentric and eccentric scales), to
detectors were positioned outside the pipeline, spaced 120◦ apart, to generate the PHDs used for training the ANNs, the F8 tally was utilized,
record the counts from a point Cs-137 source (662 keV) collimated in a and a number of starting particles (NPS = 3E8) were applied to achieve a
cone beam with a divergence angle of 8◦ . The SI, SP, and SB cards relative error of less than 5% for the entire PHD (Goorley et al., 2016).
included in the MCNP6 code were utilized for this purpose. These cards The MCNP code offers ten statistical controls to ensure the reliability of
were used to simulate the mathematical collimation of a divergent beam the estimated response, and all simulations successfully passed these
in MCNP6. The direction and divergence of the beam can be precisely controls.
defined with the help of these cards. Using the SI card to define the
angular range, the SP card to establish a uniform probability distribu­
tion, and the SB card to apply any necessary bias, a source with an 8◦ 2.6. ANN training data
divergence angle was configured. This method offers efficiency and
flexibility in defining the beam’s properties while streamlining the Artificial Neural Networks (ANNs) serve as important tools for pre­
simulation by eliminating the requirement to physically model the ge­ dicting responses in intricate systems, particularly when not all the
ometry of the collimator. The detector was experimentally validated in a required information is readily available. Therefore, ANNs were used to

6
A.C.L. Carvalho et al. Radiation Physics and Chemistry 223 (2024) 112018

Table 2
Thickness values and their uncertainties obtained from the data MCNP6 code for
68% confidence interval.
Thickness Referencea (mm) MCNP6 Thickness (cm)

Methane Oil Salty water

1 0.999 ± 1.22% 0.999 ± 2.53% 0.999 ± 2.28%

2 2.002 ± 1.08% 2.001 ± 1.28% 2.000 ± 1.15%
3 3.000 ± 1.06% 3.001 ± 0.87% 3.001 ± 0.78%
4 4.001 ± 1.05% 4.000 ± 0.66% 3.999 ± 0.59%
5 5.001 ± 1.04% 5.000 ± 0.53% 4.999 ± 0.48%
a
Conservative reference measurements were obtained from an instrument
with an uncertainty of 0.05 mm.

Table 3
Results of 5 mm concentric scale thickness and its relative error.
Scale Thickness (mm) RE (%)

h (mm) methane oil salty water methane oil salty water

0 5.0 5.1 5.2 0 2 4

Fig. 8. Relative error versus CaCO3 thickness from 1 mm to 5 mm for
5 5.0 5.1 5.2 0 2 4
different fluids.
10 5.0 5.1 5.2 0 2 4
15 5.0 5.1 5.2 0 2 4
predict scale thicknesses in both concentric and eccentric cases, irre­ 20 5.0 5.1 5.2 0 2 4
25 5.0 5.1 5.2 0 2 4
spective of the presence of fluids and the position of the eccentric scale.
30 5.0 5.1 5.2 0 2 4
For scale thickness prediction, a three-layer feed-forward Multi-Layer 35 5.0 5.1 5.1 0 2 2
Perceptron (MLP) was used, as depicted in Fig. 7. The architecture 40 5.0 5.1 5.1 0 2 2
involved a backpropagation learning algorithm, with a cross-validation 45 5.0 5.1 5.1 0 2 2
applied to prevent overtraining (Haykin, 2005). 50 5.0 5.1 5.1 0 2 2
55 5.0 5.1 5.1 0 2 2
The training dataset obtained with the MCNP6 code, consisting of 37
60 5.0 5.1 5.1 0 2 2
simulations for concentric deposition and 960 simulations for eccentric 65 5.0 5.0 5.1 0 0 2
cases, was divided into three subsets for supervised training of the ANNs: 70 5.0 5.0 5.1 0 0 2
Training, Test and Validation. These subsets constituted 70%, 20% and 75 5.0 5.0 5.1 0 0 2
80 5.0 5.0 5.1 0 0 2
10% of the data, respectively. The Pulse Height Distributions (PHDs)
85 5.0 5.0 5.1 0 0 2
input into the ANNs were adjusted for the 20–720 keV energy range, 90 5.0 5.0 5.1 0 0 2
with intervals of 10 keV per channel. 95 5.0 5.1 5.0 0 2 0
It is worth noting that at this stage, we employed two artificial neural 100 5.0 5.1 5.0 0 2 0
networks: the first network (ANN1) was designed to address concentric 105 5.0 5.1 5.0 0 2 0
110 5.0 5.1 5.0 0 2 0
scale, while the second network (ANN2) focused on eccentric scale,
115 5.0 5.1 5.0 0 2 0
since the objective is to compare the results obtained through analytical 120 244 244.0 244.0 4780 4780 4780
equations with those derived from artificial neural networks. A third 125 251.8 256.3 256.9 4936 5026 5038
neural network (ANN3) was designed to predict both concentric and
eccentric scale, utilizing a single neural network for both models. This
approach is based on the hypothesis that the internal differences in the n (
∑ )2
scale models contain sufficient information within the PHDs to enable Ri − Vpredicted
i=1
accurate predictions. R =1 −
2
n ( )2 Equation 23
∑
The performance of the proposed ANN model is evaluated using Ri − V predicted
i=1
statistical indicators, including mean absolute error (MAE) (Equation
⃒ ⃒
(19)), mean absolute percentage error (MAPE) (Equation (20)), mean ⃒Vpredicted − Vtrue ⃒
squared error (MSE) (Equation (21)), root mean square error (RMSE) RE = Equation 24
|Vtrue |
(Equation (22)), regression coefficient (R2) (Equation (23)) and relative
error (RE) (Equation (24)). Where:
n – represents the number of data;
n ⃒
1∑ ⃒ Vpredicted – predicted by ANN;
MAE = ⃒Vtrue − Vpredicted ⃒ Equation 19
n i=1 Pi – average of predicted values;
Vtrue – true or exact value.
n ⃒ ⃒
1∑ ⃒Vtrue − Vpredicted ⃒
MAPE = ⃒ ⃒ × 100% Equation 20
n i=1 ⃒ Vtrue ⃒
3. Results

n
1∑ ( )2 Scale thicknesses for concentric deposition geometry were deter­
MSE = Vtrue − Vpredicted Equation 21
n i=1 mined by measuring the transmitted flux passing through the pipe at a
height of h = 0 mm, as discussed in section 2.2. Fig. 8 illustrates the
[
n
]1/2 results for CaCO3 scale thickness ranging from of 1–5 mm with methane,
1∑ ( )2
RMSE = Vtrue − Vpredicted Equation 22 salty water, and oil inside the pipe, considering a detection height of h =
n i=1
0 mm. These results indicate that the relative error was less than 5% for
all three fluids. Consequently, a thickness of 5 mm was selected to
conduct the pipe scanning.

7
A.C.L. Carvalho et al. Radiation Physics and Chemistry 223 (2024) 112018

Table 4 Table 6
Results of 48 mm concentric scale thickness and its relative error. Results of 5 mm eccentric scale thickness and its relative error.
Scale Thickness (mm) RE (%) Position Scale Thickness (mm) RE (%)

h (mm) methane oil salty water methane oil salty water methane oil salty water methane oil salty water

0 48.1 48.4 48.1 0.2 0.8 0.2 0◦ 5.0 5.1 5.2 0 2 4

5 48.1 48.4 48.1 0.2 0.8 0.2 90◦ 5.1 5.2 5.6 2 4 12
10 48.1 48.4 48.1 0.2 0.8 0.2 120◦ 5.0 5.1 5.5 0 2 10
15 48.1 48.4 48.1 0.2 0.8 0.2
20 48.1 48.4 48.1 0.2 0.8 0.2
25 48.1 48.4 48.1 0.2 0.8 0.2
30 48.1 48.4 48.1 0.2 0.8 0.2 Table 7
35 48.1 48.4 48.1 0.2 0.8 0.2 Results of 50 mm eccentric scale thickness and its relative error.
40 48.1 48.4 48.1 0.2 0.8 0.2
Position Scale Thickness (mm) RE (%)
45 48.1 48.4 48.1 0.2 0.8 0.2
50 48.1 48.4 48.1 0.2 0.8 0.2 methane oil salty water methane oil salty water
55 48.1 48.3 48.1 0.2 0.6 0.2
60 48.1 48.3 48.1 0.2 0.6 0.2 0◦ 50.1 50.2 50.3 0.2 0.4 0.6
65 48.1 48.3 48.1 0.2 0.6 0.2 90◦ 56.3 56.6 56.8 12.6 13.2 13.6
70 48.0 48.2 48.1 0.0 0.4 0.2 120◦ 54.9 54.9 55.1 9.8 9.8 10.2
75 48.0 48.1 48.1 0.0 0.2 0.2
80 204.0 204.0 204.0 325.0 325.0 325.0
85 209.0 209.0 209.0 335.4 335.4 335.4
Table 8
90 214.0 214.0 214.0 345.8 345.8 345.8
95 219.0 219.0 219.0 356.3 356.3 356.3 Results of 103 mm eccentric scale thickness and its relative error.
100 224.0 224.0 224.0 366.7 366.7 366.7 Position Scale Thickness (mm) RE (%)
105 229.0 229.0 229.0 377.1 377.1 377.1
110 234.0 234.0 234.0 387.5 387.5 387.5 methane oil salty water methane oil salty water
115 239.0 239.0 239.0 397.9 397.9 397.9 0◦ 103.2 103.2 103.3 0.2 0.2 0.3
120 244.0 244.0 244.0 408.3 408.3 408.3 90◦ 146.1 146.2 146.2 41.8 41.9 41.9
125 251.8 251.8 251.8 424.6 424.6 424.6 120◦ 133.8 133.9 134.0 29.9 30.0 30.1

methane inside the pipe has no influence, unlike salty water, which,
Table 5
Results of 96 mm concentric scale thickness and its relative error. being the densest among the analyzed fluids, significantly affects the
counts recorded on the detector surface and increases the relative error
Scale Thickness (mm) RE (%)
(RE) (%) values. Additionally, the results for h = 120 mm and h = 125
h (mm) methane oil salty water methane oil salty water mm are meaningless since, at these positions, there is no interaction with
0 96.3 96.6 96.1 0.3 0.6 0.1 the fluids.
5 96.3 96.6 96.1 0.3 0.6 0.1 Tables 4 and 5 present the results for scale thickness of 48 and 96
10 96.3 96.6 96.1 0.3 0.6 0.1 mm, respectively. From these findings, it is apparent that the analytical
15 96.3 96.6 96.1 0.3 0.6 0.1
equations can be applied to different scale thickness values. However, it
20 96.3 96.5 96.1 0.3 0.5 0.1
25 96.2 96.3 96.0 0.2 0.3 0.0 is important to note that the results obtained at h = 80 mm for 48 mm
30 154.0 154.0 154.0 60.4 60.4 60.4 scale and at h = 30 mm for 96 mm scale do not accurately represent the
35 159.0 159.0 159.0 65.6 65.6 65.6 respective thickness of the concentric scale. At these points, there is no
40 164.0 164.0 164.0 70.8 70.8 70.8
interaction with the fluids; hence, the analytical equations are not
45 169.0 169.0 169.0 76.0 76.0 76.0
50 174.0 174.0 174.0 81.3 81.3 81.3 applicable under these conditions. Additionally, it is observed that the
55 179.0 179.0 179.0 86.5 86.5 86.5 analytical equations have the capability to calculate scale irrespective of
60 184.0 184.0 184.0 91.7 91.7 91.7 the fluid, demonstrating a maximum error of 4% for salty water, which
65 189.0 189.0 189.0 96.9 96.9 96.9 possesses the highest density, when considering the smallest scale
70 194.0 194.0 194.0 102.1 102.1 102.1
thickness of 5 mm. In other words, the equations need to be adjusted for
75 199.0 199.0 199.0 107.3 107.3 107.3
80 204.0 204.0 204.0 112.5 112.5 112.5 the fluid. This further reinforces the reason for using artificial neural
85 209.0 209.0 209.0 117.7 117.7 117.7 networks.
90 214.0 214.0 214.0 122.9 122.9 122.9 The same procedure was applied for the eccentric deposition. Ta­
95 219.0 219.0 219.0 128.1 128.1 128.1
bles 6, 7 and 8 showcase the results for the maximum eccentric scale
100 224.0 224.0 224.0 133.3 133.3 133.3
105 229.0 229.0 229.0 138.5 138.5 138.5
thickness at 0◦ , 90◦ , and 120◦ positions inside the pipeline (see Fig. 4). It
110 234.0 234.0 234.0 143.8 143.8 143.8 is evident from the low relative error values that the analytical equation
115 239.0 239.0 239.0 149.0 149.0 149.0 (see Equation (5)) can effectively determine the scale thickness when
120 244.0 244.0 244.0 154.2 154.2 154.2 maximum scale thickness is positioned at 0◦ . However, altering the
125 251.8 251.8 251.8 162.3 162.3 162.3
relative position of the maximum scale thickness within the pipe from
0◦ to 90◦ or 120◦ markedly increased the relative errors, particularly for
The uncertainties in the scale thickness, considering the investigated higher scale thickness values. Furthermore, the impact of fluid density
fluids to obtain the minimum detectable thickness, were calculated on the scale calculations was apparent in both eccentric and concentric
using Equation (17) and are presented in Table 2. models.
The most significant contributors to the uncertainty calculation of It is important to highlight that Equation (4), initially developed for
the concentric scale are the uncertainties in the pipe thickness and the I/ scales positioned at 0◦ , was also used for scales at 90◦ and 120◦ . This is
I0 values. evident from the high relative error values presented in Table 6, 7, and 8,
Table 3 presents the results of the concentric scale thickness calcu­ underscoring the necessity of determining the precise position of the
lated from Equation (5), considering a 5 mm scale. It is evident that the eccentric scale inside the pipe or conducting the pipeline scanning.
fluid density directly impacts the results. For example, the presence of The training and test sets for the ANN were created using MCNP6 to
conduct Monte Carlo simulations as the initial step in this investigation.

8
A.C.L. Carvalho et al. Radiation Physics and Chemistry 223 (2024) 112018

Fig. 9. PHDs obtained by the MCNP6 code for the beam scattered on the detectors for different scale values. For concentric scale: a) 0.6 cm; b) 5.2 cm; c) 10.8 cm. For
eccentric scale: d) 0.6 cm; e)5.4 cm; f) 10.8 cm.

As an example, Fig. 9 displays some scattered beam measurements that training for concentric and eccentric scales, respectively, which pro­
provided the pulse height distributions (PHDs). The PHDs were cate­ vided optimal results for the ANNs. The input data were normalized to
gorized within an energy range of 20–720 keV. The variations in the the range of [− 1,1]. The hidden layer comprises three independent
PHDs enabled the proposed ANN model to correlate with the maximum modules with the option to utilize activation functions and varying
scale thickness (see Table 8). number of neurons.
Tables 9 and 10 outline the parameters employed for the ANN Scale thicknesses for eccentric deposition geometry were determined

9
A.C.L. Carvalho et al. Radiation Physics and Chemistry 223 (2024) 112018

Table 9 particularly in eccentric scales, are smaller than those presented by the
Parameters applied for the ANN1 training considering concentric scales. artificial neural network. It is important to highlight that knowing the
Parameters Layers position of the eccentric scale is to utilize the appropriate analytical
equations. However, in dynamic applications involving steel tubes,
Input Hidden Output
determining the exact scale position is not feasible. Employing a single
Activation – Gaussian Tanh15 Gaussian Logistic equation, irrespective of the scale’s position, leads to significantly
function Complement
Neurons 167 34 32 32 1
inferior results compared to those achieved by the networks, as depicted
in Table 6, 7 and 8.
The two scale models were separated into concentric (ANN1) and
eccentric (ANN2) configurations, each using a dedicated neural
Table 10
network. This approach was adopted for theoretical validation with
Parameters applied for the ANN2 training considering eccentric scales.
analytical equations. However, since the internal differences in the scale
Parameters Layers are reflected in the PHDs recorded by the detectors, as shown in Fig. 9, a
Input Hidden Output new approach was implemented using both concentric and eccentric
Activation – Gaussian Tanh15 Gaussian Logistic scale data within a single neural network (ANN3). In this step, the
function Complement number of neurons in ANN3 was calculated based on the MAE and RMSE
Neurons 167 38 38 38 1 metrics. The number of neurons varied from 5 to 90. Fig. 11 presents the
results for MAE and RMSE. It can be noted that the network model with
55 neurons in the three intermediate layers presented the best metrics.
using a three NaI(Tl) detectors array and a single Cs-137 point source, as
The results for ANN3 are presented in Table 13.
discussed in Section 2.5.
For the concentric scale (ANN1), although there was a slight
Fig. 10 presents the ANN results to concentric (ANN1) and eccentric
(ANN2) scales thickness, considering all data (training, test, and vali­
dation), along with linear fitting equations obtained through the least- Table 11
square procedure and the regression coefficient (R2). It is evident that ANN processed data for concentric and eccentric models.
there is good convergence for both types of deposition. Relative error Scale Thickness
Analyzing the ANN responses for all data, it is possible to verify their Concentric (ANN1) Eccentric (ANN2)
good convergence in the case of concentric scale (ANN1). This is
<5% 94.60 91.25
underscored by applying of the least squares method for linear fitting, 5%–10% 2.70 5.42
resulting in a correlation factor (R2) of 0.9999. For the eccentric scale 10–20% 0 2.40
(ANN2), the linear fitting using the least squares method yielded an R2 of 20–30% 2.70 0.31
0.9995, indicating excellent convergence of the ANN for all data. >30% 0 0.62
R2 0.9999 0.9995
The ANN data concerning the concentric and eccentric scales, as
presented in Table 11, confirmed that in both cases, the ANNs success­
fully predicted the scale thicknesses within a 5% relative error in 94.60
% of the concentric cases and 91.25% of the eccentric cases, irrespective Table 12
Analysis of ANNs prediction using mathematical metrics.
of their position inside the pipeline.
The mean absolute error (MAE), mean squared error (MSE), root- Metric Patterns for prediction of scale thickness
mean-squared error (RMSE), and mean relative error (MRE) for all Concentric Scale (ANN1) Eccentric Scale (ANN2)
patterns and the Validation subset were employed to validate the
All (37) Validation set (3) All (960) Validation set (96)
generalization capability of the ANNs in predicting both concentric and
MAE 0.0282 0.0616 0.0851 0.1286
eccentric scale thicknesses. The results, outlined in Table 12, underscore
MSE 0.0041 0.0094 0.0227 0.0347
the ANNs’ ability to predict scale thickness for untrained patterns, RMSE 0.0644 0.0971 0.1505 0.1863
thereby highlighting their effectiveness, particularly in eccentric cases. MRE 1.19 0.48 2.00 4.45
The relative errors generated by the analytical equations,

Fig. 10. ANN results for models: a) concentric (ANN1); b) eccentric (ANN2).

10
A.C.L. Carvalho et al. Radiation Physics and Chemistry 223 (2024) 112018

Fig. 11. –Metrics used to evaluate the number of ANN3 neurons.

outcomes revealed that the networks successfully predicted over 96 % of

Table 13 the concentric and eccentric cases of CaCO3 scale thickness with a
ANN processed ALL data for ANN3.
relative error within 10%. It is worth noting that the complex detection
Relative error Scale Thickness geometry used for eccentric scale detection was also used for concentric
<5% 93.98 scale. However, a simpler geometry with only one detector might suf­
5%–10% 3.41 fice, given the symmetry of the concentric scale.
10–20% 1.41 This methodology demonstrated efficacy regardless of prior knowl­
20–30% 0.60
>30% 0.60
edge about the scale’s location inside the pipe and the presence of the
R2 0.9996 fluids, provided that ANNs are properly trained. It is noteworthy to
highlight the independence of results in eccentric models from the po­
sition of maximum thickness inside the pipeline. While this paper
decrease in errors of up to 5%, this was balanced by a reduction in errors evaluated the potential for predicting scale in mathematical simulation
in the 20–30% range. For eccentric scale, the results showed a small studies within a specific geometry, real-world applications require a
improvement. Based on these findings, it is feasible to use a single neural precise characterization of the materials to feed the MCNP6 code. In
network (ANN3) to model both types of scale effectively. addition, it is essential to assess the feasibility of installing three de­
tectors and a radiation source around the pipe. Overall, this study holds
4. Conclusions promise for applications in offshore industry systems, offering a means
to predict CaCO3 scale thickness in pipelines, which can aid in the
This study aimed to determine the thickness of CaCO3 scales within a evaluation of scale removal techniques and the utilization of inhibitors.
pipe, simulating conditions encountered in the offshore industry. In the
initial phase, the simulated model was validated using analytical CRediT authorship contribution statement
equations derived from gamma densitometry and trigonometric re­
lations for concentric CaCO3 scales with thicknesses of 5 mm, 48 mm, A.C.L. Carvalho: Writing – original draft, Validation. W.L. Salgado:
and 96 mm. Results indicated no influence of the gaseous phase. How­ Formal analysis, Data curation. R.S.de F. Dam: Writing – original draft,
ever, the presence of salty water and oil inside the pipe showed an in­ Supervision. C.C. Conti: Writing – review & editing, Data curation. C.M.
fluence due to their higher densities. Nonetheless, all relative errors Salgado: Project administration, Methodology, Conceptualization.
remained below 5 %. In eccentric cases, the analytical equation was
utilized to calculate the maximum CaCO3 thickness for positions of 0◦ ,
90◦ , and 120◦ within the pipeline, corresponding to scale thicknesses of Declaration of competing interest
5 mm, 50 mm, and 103 mm, respectively. Satisfactory results were
achieved only for the 0◦ position, primarily because the equation The authors declare that they have no known competing financial
considered the scale’s position at 0◦ , resulting in relative errors within interests or personal relationships that could have appeared to influence
5%. However, for the 90◦ and 120◦ positions, relative errors were the work reported in this paper.
higher, ranging from 9.8% to 41.9%. Similar to results observed with the
concentric scale, the influence of the higher fluid density was evident in Data availability
these cases too.
The second part of the study focused on predicting the scale thickness Data will be made available on request.
for both concentric and eccentric depositions within a three-phase flow
regime comprising oil, methane gas, and salty water using ANNs. For Acknowledgments
this purpose, a simulation of a three NaI(Tl) detectors array was con­
ducted with the MCNP6 code, and the resulting PHDs were utilized to The authors gratefully acknowledge the scholarships from Coor­
create a dataset for training and testing the ANNs. In both types of denação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil
deposition, a three-layer feed-forward (MLP) network was employed for (CAPES) (Finance Code 001) and Conselho Nacional de Desenvolvi­
scale prediction, utilizing a backpropagation learning algorithm in mento Científico e Tecnológico (CNPq). The authors also thank the
conjunction with cross-validation for improve model evaluation. The Instituto de Engenharia Nuclear (IEN) from Comissão Nacional de

11
A.C.L. Carvalho et al. Radiation Physics and Chemistry 223 (2024) 112018

Energia Nuclear (CNEN). McConn Jr., R.J., Gesh, C.J., Pagh, R.T., Rucker, R.A., Williams, III R.G., 2011.
Compendium of material composition data for radiation transport modeling,
radiation portal monitor project, pacific northwest. National Laboratory. PIET-
References 43741-TM-963 PNNL-15870 Rev. 1.
Mi, Y., Tsoukalas, L.H., Ishii, M., Xiao, Z., 1996. Hybrid fuzzy-neural flow identification
Achmad, B., Hussein, E.M.A., 2004. An X-ray Compton scatter method for density methodology. In: 5th International Conference on Fuzzy-Systems, 1996. New
measurement at a point within an object. Appl. Radiat. Isot. 60 (6), 805–814. Orleans, L.A.
Affonso, R.R.W., Dam, R.S.F., Salgado, W.L., Silva, A.X., Salgado, C.M., 2020. Flow Mitchell, R.W., Grist, D.M., Boyle, M.J., 1980. Chemical treatments associated with
regime and volume fraction identification using nuclear techniques, artificial neural North Sea projects. J. Petrol. Technol. 32 (5), 904–922.
networks and computational fluid dynamics. Appl. Radiat. Isot. 159 https://doi.org/ Moghadasi, J., Jamialahmadi, M., Müller-Steinhagen, H., Sharif, A., Izadpanah, M.,
10.1016/j.apradiso.2020.109103. Motaei, E., Barati, R., 2002. Formation damage in Iranian oil fields. In: SPE
Aggarwal, C.C., 2018. Neural Networks and Deep Learning. Springer, 978-3-319-94462- International Symposium and Exhibition on Formation Damage Control. Lafayette,
3. Louisiana. https://doi.org/10.2118/73781-MS. February 2002.
Allaga, D.A., Wu, G., Sharma, M.M.M., Lake, L.W., 1992. Barium and calcium sulfate Olajire, A.A., 2015. A review of oilfield mineral scale deposits management technology
precipitation and migration inside sand packs. SPE Form. Eval. 7, 79–86. for oil and gas production. J. Petrol. Sci. Eng. 135, 723–737.
Bertran, C.A., Ziglio, C.M., Damos, F.S; Loh, W., 2002. Cinética de formação de Oliveira, D.F., Nascimento, J.R., Marinho, C.A., Lopes, R.T., 2015. Gamma transmission
incrustações de CaCO3 e BaSO4 e o efeito de inibidores. O uso de Microbalança de system for detection of scale in oil exploration pipelines. In: Nuclear Instruments and
Quartzo (QCM) como detector. In: 25ª Reunião Anual da Sociedade Brasileira de Methods in Physics Research Section A: Accelerators, vol784. Spectrometers,
Química – SBQ. Poços de Caldas, MG. Maio, 2002. Detectors and Associated Equipment, pp. 616–620.
Bishop, C.M., James, G.D., 1992. Analysis of multiphase flows using dual-energy gamma Read, P.A., Rigen, J.K., 1982. The use of laboratory tests to evaluate scaling problems
densitometry and neural networks. Nucl. Instrum. Methods A, 327–580. during water injection. In: 6th International Symposium on Oilfield and Geothermal
Cai, S., Toral, H., Qiu, J., Archer, J.S., 1994. Neural network based objective flow regime Chemistry. Dallas, Texas. https://doi.org/10.2118/10593-MS. January 1982.
identification in air-water two phase flow. Can. J. Chem. Eng. 72 (3), 440–445. Roshani, G.H., Feghhi, S.A.H., Adineh-Vand, A., Khorsandi, M., 2013. Application of
Candeias, J.P., Oliveira M., D. Anjos, Lopes, R.T., 2014. Scale analysis using X-ray adaptive neuro-fuzzy inference system in prediction of fluid density for a gamma-ray
microfluorescence and computed radiography. Radiat. Phys. Chem. 95, 408–411. densitometer in petroleum products monitoring. Measurement 46 (9), 3276–3281.
Dam, R.S.F., Salgado, W.L., Schirru, R., Salgado, C.M., 2021. Application of radioactive Roshani, G.H., Karami, A., Salehizadeh, A., Nazemi, E., 2017. The capability of radial
particle tracking and an artificial neural network to calculating the flow rate in a basis function to forecast the volume fractions of the annular three-phase flow of gas-
two-phase (oil–water) stratified flow regime. Appl. Radiat. Isot. 180 https://doi.org/ oil-water. Appl. Radiat. Isot. 129, 156–162.
10.1016/j.apradiso.2021.110061. Roshani, G.H., Karami, A., Khazaei, A., Olfateh, A., Nazemi, E., Omidi, M., 2018.
Dam, R.S.F., Salgado, W.L., Schirru, R., Salgado, C.M., 2022. Application of deep neural Optimization of radioactive sources to achieve the highest precision in three-phase
network and gamma-ray scattering in eccentric scale calculation regardless of the flow meters using Jaya algorithm. Appl. Radiat. Isot. 139, 256–265.
fluids volume fraction inside a pipeline. Appl. Radiat. Isot. 188 https://doi.org/ Salgado, C.M., Brandão, L.E.B., Schirru, R., Pereira, C.M.N.A., Silva, A.X., Ramos, R.,
10.1016/j.apradiso.2022.110353. 2009. Prediction of volume fractions in three-phase flows using nuclear technique
Diebes, E.F., Jessen, F.W., 1967. Carbonate scale formation on steel surfaces. In: 18th and artificial neural network. Appl. Radiat. Isot. 67, 1812–1818.
Annual Technical Meeting, The Petroleum Society of CIM. Banff, Alberta. Salgado, C.M., Pereira, C.M.N.A., Schirru, R., Brandão, L.E.B., 2010. Flow regime
Goorley, T., James, M., Booth, T., Brown, F., Bull, J., Cox, L.J., Durkee, J., Elson, J., identification and volume fraction prediction in multiphase flows by means of
Fensin, M., Forster, R.A., Hendricks, J., Hughes, H.G., Johns, R., Kiedrowski, B., gamma-ray attenuation and artificial neural networks. Prog. Nucl. Energy 52 (6),
Martz, R., Mashnik, S., McKinney, G., Pelowitz, D., Prael, R., Sweezy, J., Waters, L., 555–562.
Wilcox, T., Zukaitis, T., 2016. Features of MCNP6. Ann. Nucl. Energy 87, 772–783. Salgado, C.M., Brandão, L.E.B., Schirru, R., Pereira, C.M.N.A., Conti, C.C., 2012.
Haykin, S., 2005. Neural Networks: A Comprehensive Foundation [S. l.], 9a. Pearson Validation of a NaI(Tl) detector’s model developed with MCNPX code. Prog. Nucl.
Education. ISBN 81-7808-300-0. Energy 59, 19–25.
Hubbell, J.H., Seltzer, S.M., 1997. X-Ray Mass Attenuation Coefficients. National Salgado, C.M., Brandão, L.E.B., Conti, C.C., Salgado, W.L., 2016. Density prediction for
Institute of Standards and Technology, Gaithersburg, MD [Online] Available: petroleum and derivatives by gamma-ray attenuation and artificial neural networks.
https://www.nist.gov/pml/x-ray-mass-attenuation-coefficients. Last Update to Data Appl. Radiat. Isot. 116, 143–149.
Content: July 2004, NISTIR 5632. Salgado, W.L., Sophia, de F., Dam, R., Teixeira, T.P., Conti, C.C., Salgado, C.M., 2020.
Hussein, E.M.A., Han, P., 1995. Phase volume-fraction measurement in oil-water-gas Application of artificial intelligence in scale thickness prediction on offshore
flow using fast neutrons. Nuclear Geophysics 9 (3), 229–234. petroleum using a gamma-ray densitometer. Radiat. Phys. Chem. 168 https://doi.
IAEA - International Atomic Energy Agency, 2013. Radiotracer Generators for Industrial org/10.1016/j.radphyschem.2019.108549.
Applications. Radiation Technology Series, n.5. https://www-pub.iaea.org/MTCD/p Salgado, C.M., Salgado, W.L., Dam, R.S.F., Conti, C.C., 2021a. Calculation of scales in oil
ublications/PDF/Pub1579_web.pdf. pipeline using gamma-ray scattering and artificial intelligence. Measurement 179.
IAEA - International Atomic Energy Agency, 2020. Industrial Applications of Sealed https://doi.org/10.1016/j.measurement.2021.109455.
Radioactive Sources. IAEA-TECDOC-1925. https://www-pub.iaea.org/MTCD/pu Salgado, W.L., Dam, R.S.F., Salgado, C.M., Puertas, E.J.A., Silva, A.X., 2021b. System to
blications/PDF/TE-1925web.pdf. detect the interface region and identify products transported by polyducts using
Islami Rad, S.Z., Gholipour-Peyvandi, R., Gharaghani pour, H., 2022. Evaluation of artificial intelligence and gamma radiation through MCNP6 code. Measurement 185.
nuclear data analysis techniques for volume fraction prediction in the flow meter. https://doi.org/10.1016/j.measurement.2021.110093.
Radiochim. Acta 111. https://doi.org/10.1515/ract-2022-0043. Salgado, W.L., Dam, R.S.F., Ramos, L.L., Xavier, A.S., Conti, C.C., Salgado, C.M., 2022.
JCGM - Joint Committee for Guides in Metrology, 2008. ISO/IEC Guide 98-3:2008 - Experimental and simulated methods to characterize the response of a scintillator
Uncertainty of Measurement - Part 3: Guide to the Expression of Uncertainty in detector. Radiat. Phys. Chem. 192 https://doi.org/10.1016/j.
Measurement (GUM:1995). International Organization for Standardization, Geneva, radphyschem.2021.109895.
Switzerland. Sekoguchi, K., Inoue, K., Imasaka, T., 1987. Void signal analysis and gas-liquid two phase
Kamal, M.S., Hussein, I., Mahmoud, M., Sultan, A., Saad, M.A.S., 2018. Oilfield scale flow regime determination by a statistical patter recognition method. Jpn. Soc.
formation and chemical removal: a review. J. Petrol. Sci. Eng. 171, 127–139. Mech. Eng. 30 (266), 1266–1273.
Klein, O., Nishina, Y., 1929. On the scattering of radiation by free electrons according to Sippel, R.F., Glover, D., 1964. The solution alteration of carbonate rocks, the effects of
Dirac’s new relativistic quantum dynamics. Original in Z. Phys. 52 (853) https://doi. temperature and pressure. Geochem. Cosmochim. Acta 28, 1401–1417.
org/10.1142/9789814335911_0006 reproduced here with permission from Stiff, H.A., Davis, L.E., 1952. A method for predicting the tendency of oil field waters to
Springer-Verlag. Translated from the German by Dr. Lars Bergström. deposit calcium carbonate. The American Institute of Mining. Metall. Pet. Eng. 195,
Knoll, G.F., 1999. Radiation Detection and Measurement, 2a. ed. John Wiley & Sons. 213–216.
Kumar, S., Naiya, T.K., Kumar, T., 2018. Developments in oilfield scale handling towards Tissot, B.P., Wellte, D.H., 1984. Petroleum Formation and Occurrence, vol2. Springer-
green technology - a review. J. Petrol. Sci. Eng. 169, 428–444. Verlag Berlin Heidelberg, New York Tokyo. ISBN 0-38713281-3.
Lindlof, J.C., Stoffer, K.G., 1983. A case study of seawater injection incompatibility. Triggia, A.A., Correia, C.A., Filho, C.V., Xavier, J.A.D., Machado, J.C.V., Tomas, J.E.,
J. Petrol. Technol. 35 (7), 1256–1262. Filho, J.E.S., Paula, J.L., Rossi, N.C.M., Pitombo, N.E.S., Gouvea, P.C.V.M.,
Lopes, J.M., Mansur, C.R.E., Neves, G.B.M., Lucas, E.F., 2003. Avaliação de aditivos Carvalho, R.S., Barragan, R.V., 2004. Fundamentos da Engenharia de Petróleo, 2a.
inibidores de incrustação gerada a partir da precipitação de CACO3. In: 2o Congresso ed. Editora Interciência, Rio de Janeiro, ISBN 85-7193-046-5.
Brasileiro de P&D em Petróleo & Gás. Watkinson, A.P., Martinez, O., 1975. Scaling of heat exchanger tubes by calcium
carbonate. J. Heat Tran. 97 (4), 504–508.

12