materials

Article
3D Model of Carbon Diffusion during Diffusional
Phase Transformations
Łukasz Łach * and Dmytro Svyetlichnyy

AGH University of Krakow, Faculty of Metals Engineering and Industrial Computer Science, al. Mickiewicza 30,
30-059 Krakow, Poland; svetlich@metal.agh.edu.pl
* Correspondence: lach@agh.edu.pl

Abstract: The microstructure plays a crucial role in determining the properties of metallic materials,
in terms of both their strength and functionality in various conditions. In the context of the formation
of microstructure, phase transformations that occur in materials are highly significant. These are
processes during which the structure of a material undergoes changes, most commonly as a result of
variations in temperature, pressure, or chemical composition. The study of phase transformations is
a broad and rapidly evolving research area that encompasses both experimental investigations and
modeling studies. A foundational understanding of carbon diffusion and phase transformations in
materials science is essential for comprehending the behavior of materials under different conditions.
This understanding forms the basis for the development and optimization of materials with desired
properties. The aim of this paper is to create a three-dimensional model for carbon diffusion in the
context of modeling diffusional phase transformations occurring in carbon steels. The proposed
model relies on the utilization of the LBM (Lattice Boltzmann Method) and CUDA architecture.
The resultant carbon diffusion model is intricately linked with a microstructure evolution model
grounded in FCA (Frontal Cellular Automata). This manuscript provides a concise overview of the
LBM and the FCA method. It outlines the structure of the developed three-dimensional model for
carbon diffusion, details its correlation with the microstructure evolution model, and presents the
developed algorithm for simulating carbon diffusion. Demonstrative examples of simulation results,
illustrating the growth of the emerging phase and affected by various model parameters within
particular planes of the 3D calculation domain, are also presented.

Citation: Łach, Ł.; Svyetlichnyy, D. 3D Keywords: phase transformation; carbon diffusion; frontal cellular automata; lattice Boltzmann
Model of Carbon Diffusion during method; CUDA
Diffusional Phase Transformations.
Materials 2024, 17, 674. https://
doi.org/10.3390/ma17030674
1. Introduction
Academic Editor: Alexander
Yu Churyumov The microstructure plays a crucial role [1] in forming the mechanical properties of
materials. Materials engineers strive to manipulate the microstructure [2] through processes
Received: 14 December 2023 such as heat treatment, mechanical processing, and other methods to achieve the desired
Revised: 21 January 2024
material characteristics. In cases of metals, ceramics, polymers, and other materials, the
Accepted: 25 January 2024
microstructure significantly influences their strength [3], hardness, plasticity, brittleness,
Published: 30 January 2024
and many other mechanical properties. One of the fundamental phenomena that affects
the development of the microstructure in materials is phase transformation [4].
Phase transformations in materials depend on a material’s chemical composition and
Copyright: © 2024 by the authors.
refer to changes in the state of matter or internal structure of a material in response [5] to
Licensee MDPI, Basel, Switzerland. changes in conditions, such as temperature and pressure.
This article is an open access article Phase transformations in steels can be divided into two main groups [6]: diffusional
distributed under the terms and transformations [7] and diffusionless transformations [8]. Diffusionless transformations are
conditions of the Creative Commons usually much faster [9] than diffusional transformations, and they occur under extreme
Attribution (CC BY) license (https:// cooling (and heating) or deformation conditions [10]. On the other hand, diffusional
creativecommons.org/licenses/by/ transformations require longer times (especially at low temperatures [7]), allowing for the
4.0/). diffusion of atoms and a more gradual change in the crystalline structure.

Materials 2024, 17, 674. https://doi.org/10.3390/ma17030674 https://www.mdpi.com/journal/materials

Materials 2024, 17, 674 2 of 14

Among recent works related to diffusional phase transformations, the work of

Wang et al. [11] and Liu et al. [12] are worthy of attention. Wang et al. present progress
in research on the diffusional phase transformations of Fe–C alloys under high magnetic
fields. The authors show that a high magnetic field can provide the driving force for
transformations and promote ferrite and pearlite transformations and carbide precipitation
and that determining the laws that govern the changes in physical parameters, such as
the interfacial energy and diffusion coefficients under a high magnetic field, should be the
basis of future research on diffusional phase transformations. Liu et al. published a review
on the modeling and simulations of solid-state diffusional phase transformations in metals
and alloys. In the revision, the authors noted that the traditional diffusion-controlled and
interface-controlled models fail to correctly describe the growth kinetics of the new phase
over the entire course of the transformation, and the empirical and analytical models only
provide average information on the macroscale.
The studies of carbon diffusion during phase transformations retain their significant
importance [13] in materials science, engineering, and various industrial applications.
Understanding how carbon diffuses during phase transformations is crucial to designing
and optimizing materials with specific properties. For example, in the development of
advanced steels, knowledge of carbon diffusion during phase changes helps engineers
tailor [14] the strength, hardness, and corrosion resistance of the material. Carbon diffusion
during phase transformations also provides insight into the thermodynamics and kinetics
of these processes [12]. Researchers can study the energy barriers [15] involved and the time
required for transformations [16], thereby leading to better control and predictability. In
industries such as steel manufacturing, the study of carbon diffusion is vital for optimizing
processes and reducing energy consumption. It allows for the development of more energy-
efficient methods while maintaining product quality.
Advanced computational methods [12] are very helpful in understanding carbon
diffusion during diffusional phase transformations, which is a highly complex and multi-
faceted process. Advanced computational methods provide [17] the ability to model and
simulate intricate interactions, which offer insights that are difficult to obtain by analytical
or experimental means alone. Computational methods allow researchers to obtain quanti-
tative results on the diffusion of carbon atoms, such as concentration profiles [18], diffusion
coefficients [19], and diffusion rates [20]. These insights are essential for making accurate
predictions and optimizing material properties. Topics related to carbon diffusion in steels
are important when taking into account different model studies. Lee et al., for example,
proposed a simple and computationally efficient equation for carbon diffusion in austenite
in iron–carbon alloys, as well as in steels alloyed with a variety of elements [21]. The
authors of the work presented different equations for the diffusivity of carbon in austenite
and developed empirical equations to predict carbon diffusivity in austenite as a func-
tion of chemical composition and temperature for the iron–carbon-based multicomponent
alloy steels.
Various phenomena and processes in engineering can be modeled using different
numerical methods. The choice of a specific method depends on the nature of the prob-
lem, the availability of data, the precision of the expected result, and the computational
resources [22,23]. Often, various numerical methods are combined to achieve a more com-
prehensive analysis of a given phenomenon [24]. Different methods can also be employed
to model phase transformations.
Recently, Yamanaka [19] presented recent advances in phase-field modeling and simu-
lation of solid-state phase transformations in iron and steel, with particular attention paid
to the modeling of austenite-to-ferrite, pearlitic, bainitic, and martensitic transformations.
The author emphasized that the phase-field method can be applied to various solid-state
phase transformations in iron and steel because it can simulate multi-physics phenomena,
such as the migration of interfaces, the diffusion of solute atoms, and the evolution of the
stress–strain fields during phase transformations. The author also noted that the current
phase-field models are generally unable to easily provide quantitative simulation results
Materials 2024, 17, 674 3 of 14

for the solid-state phase transformations taking place in multicomponent steels that contain
substitutional alloying elements. Lahiri presented a review on phase-field modeling of
phase transformations in multicomponent alloys [25], where he noticed that, in view of the
versatility and quantitative accuracy of the discussed PF models, it can be anticipated that
simulations of microstructure evolution using such quantitative PF models will continue
to accelerate the design and development of novel multicomponent alloys in the future.
Lv et al. [26] found that numerical uncertainty and parameter sensitivity have a major
impact on the simulation results in all kinds of phase-field models.
FCA and the LBM are highly effective approaches for simulating various phenom-
ena and processes in the field of materials science. Recently, Svyetlichnyy analyzed the
computational time complexity of Frontal Cellular Automata (FCAs) and compared them
with Classic Cellular Automata (CCAs) [27], including phase transformation [28]. FCA
has a very wide range of applications in the context of modeling various processes and
phenomena occurring in materials. This method, among others, has been used successfully
for the modeling of recrystallization [29], additive manufacturing [29,30], and rolling [31].
Recently, parallel computing modeling is also being realized on a larger scale; for example,
Zhu et al. [32] realized the three-dimensional multi-phase-field simulation of a eutectoid
alloy based on an OpenCL parallel.
This paper presents a 3D model that focuses on the diffusion of carbon in diffusion
phase transformations. This model is one of the most important components of a sophisti-
cated system designed to model such transformations. Built upon two modeling methods,
FCA and the LBM, the 3D model incorporates the utilization of hybrid computing systems
(GPU and CPU) for calculations. Fundamental details regarding the FCA method and
the LBM are presented. The structure and primary assumptions of the model are also
demonstrated. This paper presents some modeling results of the transformation obtained
by this model. Parallel computing is performed on NVIDIA graphics cards with CUDA
architecture and with visualization using Open GL.

2. Lattice Boltzmann Method

The LBM originated from a variant of the cellular automata method known as the
lattice gas automaton (LGA) [33]. The basis of the method is to solve the Boltzmann
transport equation:
∂f ∂f F ∂f
+ ·ξ+ · =Ω (1)
∂t ∂x m ∂e
where f (x,u,t) represents the particle distribution function, x and ξ denote phase space
variables (coordinate and velocity), t signifies time, F stands for external force (e.g., gravity),
m is the mass, u represents macroscopic velocity, Ω is the collision operator, and e represents
lattice velocities.
Particles are distributed according to velocities and directions, as follows:
!
ρ ( ξ − u)2
f = exp − (2)
(2πRT ) D/2 2RT

where R is the gas constant, D represents the space dimension, ρ signifies gas density, and
T denotes temperature.
Along with the discretization of space and time, the LBM uses the discretization of
velocities by dividing them into a limited number of directions and setting the velocities
according to the lattice space. The velocities ξ are discretized to a set of such discrete
velocities e. Equation (1) is approximated for characteristic velocities e determined by the
grid and by directions using trapezoidal integration for spatial variables and an explicit
Euler scheme for time derivatives. The result of such a discrete approximation is the lattice
Boltzmann equation (LBE), which presents an evolution of components of the distribution
function as follows:
f i (x + ei , t + 1) = f i (x, t) + Ωi (3)
Materials 2024, 17, 674 4 of 14

Equation (3) is the foundation of the LBM.

Discretization of two- or three-dimensional space consists of imposing a regular orthog-
onal grid on a uniform lattice of dimensionless length (the distance between neighboring
nodes) equal to one dx = dy = dz = 1. The parameter dt related to the dimensionless time
step is also set to unity dt = 1. A fundamental aspect of the system involves selecting a
set of velocities e, known as the velocity model, for example, the D2Q9 model, where D
indicates dimensionality (D2—two-dimensional model) and Q indicates the number of
velocities. In one-dimensional space, the primary models used are D1Q2 or D1Q3, while in
two-dimensional space, models such as D2Q4, D2Q5, or D2Q9 are commonly employed.
For three-dimensional problems, models like D3Q6, D3Q7, D3Q15, D3Q19, and others
find application.
The important element of this equation (Equation (3)) is the collision operator Ωi . In
general, this operator is a complex nonlinear integral. An approximation of this integral
can be fulfilled by linearization near a local equilibrium point. Several approximations
are developed: single-time, two-time, and multi-time relaxation schemes and a cascade
scheme. The first and simplest single-time relaxation scheme was proposed by Bhatnagar,
Gross, and Krook and was called the BGK collision operator:

1h eq
i
Ωi ( f ) = − f i (x, t) − f i (x, t) (4)
τ
where τ—relaxation time and feq —equilibrium distribution function.
The BGK collision operator (Equation (4)) contains an equilibrium distribution function
eq
f , which can take the following form:
!
eq u · ei ( u · ei ) 2 u · u
f i (x, t) = wi ρ 1 + 2 + − 2 (5)
cs 2c4s 2cs

where ρ—density,
√ u—lattice velocity, cs —sound speed in the modeled fluid (for D2Q9), and
cs = 1/ 3, wi —directional weights (for example, for D2Q9: w0 = 4/9; w1 = w2 = w3 = w4 =
1/9; w5 = w6 = w7 = w8 = 1/36).
The macroscopic variables, density ρ and velocity u (through the momentum ρu)
(Equation (5)), are calculated as follows:

b
ρ= ∑ fi (6)
i =0

b
ρu = ∑ f i ei (7)
i =0

The LBE (Equation (3)) is solved in two stages: collision and streaming. The post-
collision form of the initial distribution function is then as follows:
1 h in eq
i
f iout (x, t) = f iin (x, t) − f i (x, t) − f i (x, t) + Fi (8)
τ
The streaming is a direct transfer of components of distribution functions to appropri-
ate neighboring nodes:
f iin (x + ei , t + 1) = f iout (x, t) (9)
Boundary conditions can be considered as a separate operation or in conjunction with
a streaming.
Figure 1 shows the general modeling algorithm using the LBM.
 priate neighboring nodes:

𝑓 𝒙 𝒆 ,𝑡 1 𝑓 𝒙, 𝑡 (9)

Boundary conditions can be considered as a separate operation or in conjunction with

Materials 2024, 17, 674
a streaming. 5 of 14
Figure 1 shows the general modeling algorithm using the LBM.

Figure 1. Graphical representation of main algorithm of LBM [33].

Figure 1. Graphical representation of main algorithm of LBM [33].
A detailed description of the LBM can be found elsewhere [33].
A detailed description of the LBM
3. Frontal Cellular can be found elsewhere [33].
Automata
The frontal cellular automaton is a specific type of computational algorithm based on
3. Frontal Cellular Automata
cellular automaton models. The name “frontal” comes from one of its most widespread
The frontal cellular automaton
applications, is agrain
namely, specific type during
growth of computational algorithm
the microstructure based onwhere the bound-
evolution,
cellular automaton models. The name “frontal” comes from one of its most widespread
ary of a growing grain is considered as the front of changes occurring in the structure. The
applications, namely, grain growth
main differences in FCAduring
from the microstructure
classical evolution,
CA are the reversal where the sending (the cell
of information
boundary of a growing grain is considered as the front of changes
sends information to its neighbors instead of receiving informationoccurring in thefrom
struc- the neighborhood),
ture. The main differences in FCA from classical CA are the reversal of information
the introduction of an additional state into the automaton, and the organization send- of the cells
ing (the cell sends information
into lists. As atoresult,
its neighbors
only smallinstead
part of
of receiving
all cells isinformation from the in one time step,
used in calculations
neighborhood), the introduction
every of an additional
cell is maintained only once state intoentire
in the the automaton,
simulation,and the the
timeorgan-
complexity is reduced,
ization of the cellsandinto lists. As atime
simulation result, only small
decreases part of all[27].
significantly cells is used in calculations
in one time step, everyThreecell isareas
maintained only onceduring
can be separated in the entire simulation,
grain growth the
in the timetransformation
phase com- process
plexity is reduced, and simulation
(Figure time decreases
2). Transformation significantly
does not begin in the [27].first area, and cells remain in the initial
Three areas state.
can beTheseparated during grain
transformation ends growth in the area:
in the second phasethe transformation
cells are in the pro-
final state, and this
cess (Figure 2). Transformation does not begin in the first area, and cells remain
state will not change. The third area is between the first two: it separates them, and the in the
initial state. The transformation
transformation ends occursinin
thethis
second
area. area:
Onlythe thecells
thirdare
areain isthe final state,
utilized in FCAandcalculations. Both
Materials 2024, 17, x FOR PEER REVIEW 6 of 16
this state will notthechange. Thesecond
first and third area
zonesis between
should be theexcluded
first two:from it separates them, and
the calculations in the current step
the transformation occurschanges
because in this area.
in theOnly the third
cell states areathese
within is utilized
zonesin are FCA
not calculations.
expected and cannot occur.
Both the first and second zones should be excluded from the calculations in the current
step because changes in the cell states within these zones are not expected and cannot
occur.

Figure 2. FCA zones: 1—cells in the initial state, 2—cells in the final state, 3—cells participating
Figure 2. FCA zones: 1—cells in the initial state, 2—cells in the final state, 3—cells participating in
in computations.
computations.

The use of FCA (Frontal Cellular Automaton) instead of conventional cellular autom-
ata reduces simulation time not only in 2D models but also in 3D CA, where it is more
noticeable. This is because significant spatial areas are excluded from the calculations at
each time step, and only a thin layer participates in the computations. In classical CA, each
step requires practically the same computational effort, and the time taken for its deter-
Materials 2024, 17, 674 6 of 14

The use of FCA (Frontal Cellular Automaton) instead of conventional cellular automata
reduces simulation time not only in 2D models but also in 3D CA, where it is more
noticeable. This is because significant spatial areas are excluded from the calculations
at each time step, and only a thin layer participates in the computations. In classical
CA, each step requires practically the same computational effort, and the time taken for
its determination remains unchanged throughout the entire simulation. In contrast, the
duration of one time step calculations in FCA depends on the number of participating cells,
and it exhibits significant variability, but it always remains a small fraction compared to
the duration in classical CA. Each cell in FCA participates in calculations only once, in
one time step, during the entire simulation, whereas in classical CA the cells participate in
calculations in all time steps.
Details about FCA can be found elsewhere [27].

4. 3D Carbon Diffusion Model

A multiscale model, which contains the 3D carbon diffusion model based on the LBM
and
a microstructure evolution model based on FCA, is shown in Figure 3. This multiscale
Materials 2024, 17, x FOR PEER REVIEW 7 of 16
model was adapted for parallel calculations on GPU with the CUDA architecture.

Figure3.3.AA3D
Figure 3Dmodel
modelof
ofcarbon
carbondiffusion
diffusion(based
(basedon
onLBM)
LBM)and
andits
itsintegration
integrationwith
withthe
themicrostructure
microstructure
evolution model (based on FCA) (red arrows symbolize the interaction between these models).
evolution model (based on FCA) (red arrows symbolize the interaction between these models).

Phaseseparation,
Phase separation,carbon
carbondiffusion
diffusionin
indifferent
differentphases,
phases,and
andcalculation
calculationof
ofthe
thephase
phase
fractionin
fraction incells
cellson
onthe
thephase’s
phase’sboundaries
boundariesare
arethe
theadditional
additionalelements
elementsimplemented
implementedinto
into
thestandard
the standardLBM
LBMmodeling
modelingscheme.
scheme.These
Theseelements
elementsare
areshown
shownininFigure
Figure44as
asaacommon
common
blocknamed
block named‘calculation
‘calculationofofphase
phase(carbon
(carbondiffusion)’.
diffusion)’.
The diffusion model is based on the 3D Fourier equation [34]:
 2
∂2 c ∂2 c

∂c ∂ c
=D + + (10)
∂t ∂x2 ∂y2 ∂z2

where c—carbon concentration and D—carbon diffusion coefficient.

 Figure 3. A 3D model of carbon diffusion (based on LBM) and its integration with the microstructure
evolution model (based on FCA) (red arrows symbolize the interaction between these models).

Phase separation, carbon diffusion in different phases, and calculation of the phase
Materials 2024, 17, 674 fraction in cells on the phase’s boundaries are the additional elements implemented into 7 of 14
the standard LBM modeling scheme. These elements are shown in Figure 4 as a common
block named ‘calculation of phase (carbon diffusion)’.

Figure
Materials 2024, 17, x FOR PEER 4.Figure
Carbon diffusion
REVIEW in thediffusion
4. Carbon calculation in
of phase in the LBM algorithm.
the calculation of phase Particular
in thecolors
LBMrep-
algorithm.
8 of 16 Particular colors
resent various stages of the cyclically realized algorithm.
represent various stages of the cyclically realized algorithm.
The diffusion model is based on the 3D Fourier equation [34]:
The Bhatnagar–Gross–Krook
𝜕𝑐 (BGK)
𝜕 𝑐 collision
𝜕 𝑐 𝜕 𝑐operation used in the LBM [33] (Equation (4))
is applied for the calculations. 𝐷 (10)
𝜕𝑡 𝜕𝑥 𝜕𝑦 𝜕𝑧
The equilibrium distribution function is calculated as follows:
where c—carbon concentration and D—carbon diffusion coefficient.
The Bhatnagar–Gross–Krook (BGK) collisioneq operation used in the LBM [33] (Equa-
tion (4)) is applied for the calculations. f k = wk c( x, y, z, t) (11)
The equilibrium distribution function is calculated as follows:
where c is the carbon concentration.
𝑓 𝑤 𝑐 𝑥, 𝑦, 𝑧, 𝑡 (11)
Figure 5 shows the D3Q19 velocity model [35], which is used for the simulations.
where
The c is thevelocities
lattice carbon concentration.
ei and the corresponding lattice factors for the calculation schemes
Figure 5 shows the D3Q19 velocity model [35], which is used for the simulations. The
are the following: e(0) = (0,0,0), e(1,2) = (±1,0,0), e(3,4) = (0,±1,0), e(5,6) = (0,0,±1), e(15,16,17,18)
lattice velocities ei and the corresponding lattice factors for the calculation schemes are the
(±1,±1,0),
=following: e(0) = e(0,0,0), e(1,2) = =
(11,12,13,14) (±1,0,e(3,4)
(±1,0,0), ±1), e(7,8,9,10)
= (0,±1,0), e(5,6)== (0, ±1,±e1),
(0,0,±1), and w(0) = 1/3, w(1–6) = 1/18,
(15,16,17,18) = (±1,±1,0),
we(11,12,13,14)= 1/36.
(7–18) = (±1,0,±1), e(7,8,9,10) = (0,±1,±1), and w(0) = 1/3, w(1–6) = 1/18, w(7–18) = 1/36.

Figure 5. D3Q19 LBM lattice. The numbers represent particular directions, and the colors mark the
Figure 5. D3Q19
perpendicular planes.LBM lattice. The numbers represent particular directions, and the colors mark the
perpendicular planes.
The neighborhood is one of the most important parameters of cellular automata [36].
The simulations were performed using the three-dimensional von Neumann and Moore
neighborhoods (Figure 6).
 Materials 2024, 17, 674 8 of 14

Materials
Materials 2024,
2024, 17, 17, x FOR
x FOR PEER
PEER REVIEW
REVIEW 9 of [36].
16 9 of 16
The neighborhood is one of the most important parameters of cellular automata
The simulations were performed using the three-dimensional von Neumann and Moore
neighborhoods (Figure 6).

(a)(a) (b)(b)
Figure
Figure 6. Moore
6.6.Moore (a)
(a)(a)
andand von Neumann (b) neighborhoods. The cell under consideration is shown
Figure Moore andvon
vonNeumann
Neumann(b) (b)neighborhoods.
neighborhoods.The
Thecell
cellunder
underconsideration
considerationisisshown
shown in
in burgundy and the cells in a given neighborhood are shown
in burgundy and the cells in a given neighborhood are shown in orange.in orange.
burgundy and the cells in a given neighborhood are shown in orange.
Figure
Figure
Figure 77 shows
shows
7 shows an an
an algorithm
algorithm that
that
algorithm is is
used
is used
that to model
to model
used carbon
carbon
to model diffusion
diffusion
carbon during
during
diffusion thethe trans-
trans-
during the
formation.
formation.
transformation.

Figure
Figure
Figure 7.Algorithm
7.7. Algorithm
Algorithm for
forfor simulation
simulation
simulation ofcarbon
ofof carbon
carbon diffusion.
diffusion.
diffusion.

The
TheThe algorithm
algorithm
algorithm contains
contains
contains a acycle
acycle
cycle with
withwithanan
an initial
initial
initial step.
step.
step. The
TheThecycle
cycle
cycle involves
involves
involves step
step
step nos.
nos.nos. 2–8.
2–8.2–8.
1. Initial
1. 1. Initial
Initial calculation,
calculation,
calculation, simulation,
simulation,
simulation, and and
and model
model
model parameters
parameters
parameters set:
set:
set:
- -- Lattice
Lattice
Lattice sizes:
sizes:
sizes: nx, nx,
nx,ny,ny,
ny,
and and
and nz;
nz;
nz;
- -- Nodes
Nodes coordinates:
coordinates:
Nodes coordinates: x, y, z; x,
x, y,
y, z;
z;
- -- Lattice
Lattice
Lattice lengths:
lengths:
lengths: x, x,
∆x, y,
y,∆y, z;
z;∆z;
- -- Time
Time
Time step:
step:
step: t;
t;∆t;
- -- Maximal
Maximal time
time
Maximal time steps: steps:
steps: tsteps;
tsteps;
tsteps;
- - Austenite () and ferrite ()(
Austenite ( ) and ferrite ) fractions
fractions in interface
in interface I I= 1=and
nodes:
nodes: 1 andII= 0;= 0;
- - Carbon concentration
Carbon concentration in nodes: ci;in nodes: c i;

- - Diffusion
Diffusion coefficient:
coefficient: DcfD. cf.
Materials 2024, 17, 674 9 of 14

- Austenite (γ) and ferrite (α) fractions in interface nodes: φIγ = 1 and φIα = 0;
- Carbon concentration in nodes: ci ;
- Diffusion coefficient: Dcf .
2. Calculation of basic parameters:
- Calculation of a boundary velocity v:v = ∆c; ∆c—difference in the concentration
between the interface node and the neighboring austenite nodes;
- Calculation of changes of phase fraction in interface node: ∆φI = v∆t;
- Calculation of new fraction: φIα,t = φIα,t −1 + ∆φI .
3. State checking:
(a) If φIα,t < 1, the interface state of the node remains, go to point 4;
(b) If φIα,t ≥ 1, the state of the node changes from I (interface) to α (ferrite, fraction
becomes φIα,t ′ = 1), the neighboring nodes (in the Moore neighborhood) in the
state γ (austenite) change their state to I (interface), fraction for the new nodes
in state I are calculated according to: ∆φnI = (φIα,t − 1)/numγ = φIα,nI ; numγ—
the number of nodes that changed their state from γ to I, the ∆φ surplus over
unity: ∆φoI = 1 − φIα,t−1 ; φIα, oI = 1; φIγ, oI = 0.
4. Calculation of new concentration: c = Σfi ;
5. Determination of the equilibrium distribution functions feq ;
6. Collision—calculation of fout ;
7. Streaming—determination of fin ;
8. Boundary conditions.

5. Simulation Results
A comprehensive multidimensional analysis was carried out on the three-dimensional
model of carbon diffusion. The established fundamental linkage with the microstructure
evolution model, utilizing the FCA method, enabled, among other things, the observation
of transformation rates. Selected outcomes of the three-dimensional carbon diffusion
modeling are showcased for specified parameter values within the model.
The first two of the modeling variants concern the modeling of carbon diffusion using
the developed algorithm in a system composed of one ferrite grain located in the middle
of the space. Two variants of the FCA neighborhoods were used for the simulations: von
Neumann and Moore. The growth of the grain and changes in carbon concentration were
modeled. The simulations were performed utilizing the D3Q19 LBM scheme (Figure 5).
The modeling relies on the subsequent fundamental parameters: nx × ny × nz = 128 × 128
× 128, ∆t = 1, τ = 1, ∆x = 1, ∆y = 1, ∆z = 1. Boundary conditions using the bounce-back
method were implemented at the edges. The initial concentration for all nodes in the
calculation grid was set to 0.6. Concentration in ferrite (α)—0.1, austenite (γ)—0.9, and
in interface (ciface )—0.5. The computations were executed utilizing the CUDA parallel
programming architecture on graphics cards, namely, NVIDIA GeForce GTX 1060 and
NVIDIA GeForce GTX 2080Ti (NVIDIA, Santa Clara, CA, USA).
The first variant with the von Neumann neighborhood is presented in Figure 8. It
shows, on the selected planes (x= nx /2, y = 1, 2, . . ., ny , z = 1, 2, . . ., nz ) of the 3D model
space, the growth of a ferrite grain (Figure 8a) and the carbon concentration distributions
(Figure 8b). Different blue intensities represent different concentration values, from smaller
values in light blue to larger values in dark blue.
Materials 2024, 17, x FOR PEER REVIEW 11 of 16

Materials 2024, 17, x FOR PEER REVIEW

674 1110of
of 16
14

Figure 8. FCA states (a) and carbon concentration distributions (b) at various growth stages of the
new phase
Figure utilizing
8. FCA statesthe
(a) von
andNeumann neighborhood.
carbon concentration distributions (b) at various growth stages of the
Figure 8. FCA states (a) and carbon concentration distributions (b) at various growth stages of the
new phase utilizing the von Neumann neighborhood.
new phase utilizing the von Neumann neighborhood.
Figure 9a shows the second variant with the Moore neighborhood. It presents the
grainFigure
growth 9afor the identical
shows the secondselected
variantplane as the
with the variant
Moore in the von Neumann
neighborhood. It presentsneighbor-
the grain
Figure 9a shows the second variant with the Moore neighborhood. It presents the
hood.
growth Distributions of carbon
for the identical concentrations
selected plane as theare presented
variant in thein Figure
von 9b. Theneighborhood.
Neumann grain growth
grain growth for the identical selected plane as the variant in the von Neumann neighbor-
Distributions
rate is directlyofrelated
carbontoconcentrations are presented
the carbon diffusion process,in and
Figure
one9b.
of The grain driving
the main growth forces
rate is
hood. Distributions of carbon concentrations are presented in Figure 9b. The grain growth
directly related to the carbon diffusion process, and one of the main driving
of the growth of the new phase is then the difference in carbon concentrations between forces of the
rate is directly related to the carbon diffusion process, and one of the main driving forces
growth
the of the new phase is then the difference in carbon concentrations between the phases.
phases.
of the growth of the new phase is then the difference in carbon concentrations between
the phases.

Figure 9. FCA states (a) and carbon concentration distributions (b) at various growth stages of the
Figure 9. FCA states (a) and carbon concentration distributions (b) at various growth stages of the
new phase utilizing the Moore neighborhood.
new phase utilizing the Moore neighborhood.
Figure 9. FCA states (a) and carbon concentration distributions (b) at various growth stages of the
In theutilizing
new phase next presented
the Mooremodeling results, the microstructure evolution with five ran-
neighborhood.
domly located grains was analyzed. The research was carried out for another selected plane
Materials 2024, 17, x FOR PEER REVIEW 12 of 16

Materials 2024, 17, 674 In the next presented modeling results, the microstructure evolution with five11ran- of 14
domly located grains was analyzed. The research was carried out for another selected
plane (x = 1, 2, …, nx, y = ny/2, z = 1, 2, …, nz) in the 3D modeled space. The basic parameters
used
(x = 1,for
2, the
. . ., modeling were
nx , y = ny /2, z =as1,follows:
2, . . ., nza) number
in the 3Dofmodeled
nodes—nspace.
x × ny × nz = 128 × 128 × 128,
The basic parameters
t x y = 1, z
used for the modeling were as follows: a number of nodes—nx × ny × nconditions,
= 1, = 1, = 1, the bounce-back method for boundary and the
z = 128 × 128 × 128,
Moore
∆t = 1, ∆x = 1, ∆y = 1, ∆z = 1, the bounce-back method for boundary conditions, andwas
neighborhood. The initial concentration for all nodes in the calculation grid the
set
Mooreto 0.6. Concentration
neighborhood. Theininitial
ferrite—0.1,
concentration austenite—0.9, andininthe
for all nodes interface—0.5. Figure
calculation grid 10
was set
shows the grains’ growth and the distribution of the carbon concentration
to 0.6. Concentration in ferrite—0.1, austenite—0.9, and in interface—0.5. Figure 10 shows in different
stages.
the grains’ growth and the distribution of the carbon concentration in different stages.

Figure 10. FCA states (a) and carbon concentration distributions (b) at various growth stages of the
Figure 10. FCA states (a) and carbon concentration distributions (b) at various growth stages of the
new phase grains utilizing the Moore neighborhood.
new phase grains utilizing the Moore neighborhood.
The influence of the initial carbon concentration in interface nodes ciface on grain
The influence of the initial carbon concentration in interface nodes ciface on grain
growth and carbon concentration distribution is shown in Figure 11. The size and shape
growth and carbon concentration distribution is shown in Figure 11. The size and shape
of the grains and the carbon concentration distributions are shown for the same time step
of the=grains
tstep and the carbon
400. Identical valuesconcentration distributions
as in the previous variantare shown
were usedforforthe same
the time step
foundational
tstep = 400. Identical values as in the previous variant were used for the foundational
parameters, and the same plane was taken into account. The growth of six randomly pa-
rameters, and the same plane was taken into account.
distributed grains of the new phase was considered. The growth of six randomly distrib-
uted grains of the
The initial new phase
carbon was considered.
concentration in interface nodes directly affects the transformation
rate. The difference in concentration between the interface nodes and the neighboring
austenite nodes affects the calculation of the boundary velocity. The determination of the
volume of the transformed material in the interface node is directly related to the boundary
velocity and affects the changes in the new phase fraction. Ferrite grains grow faster under
the condition of a lower ciface .
The influence of the initial carbon content cin on grain growth and carbon concentra-
tion distribution is shown in Figure 12. The size and shape of the grains and the carbon
concentration distributions are shown for the same time step tstep = 700. The base parame-
ters retained the same values as in the previous variant, and the analysis was conducted
on the same plane. The calculations were realized with the use of the von Neumann
neighborhood.
Materials
Materials 2024,
2024, 17,
17, x674
FOR PEER REVIEW 1312ofof 16
14

Materials 2024, 17, x FOR PEER REVIEW 14 of 16

Figure 11. FCA states (a) and carbon concentration distributions (b) for different values of ciface .
Figure 11. FCA states (a) and carbon concentration distributions (b) for different values of ciface.

The initial carbon concentration in interface nodes directly affects the transformation
rate. The difference in concentration between the interface nodes and the neighboring aus-
tenite nodes affects the calculation of the boundary velocity. The determination of the vol-
ume of the transformed material in the interface node is directly related to the boundary
velocity and affects the changes in the new phase fraction. Ferrite grains grow faster under
the condition of a lower ciface.
The influence of the initial carbon content cin on grain growth and carbon concentra-
tion distribution is shown in Figure 12. The size and shape of the grains and the carbon
concentration distributions are shown for the same time step tstep = 700. The base param-
eters retained the same values as in the previous variant, and the analysis was conducted
on the same plane. The calculations were realized with the use of the von Neumann neigh-
borhood.

Figure 12. FCA states (a) and carbon concentration distributions (b) for different values of cin .
Figure 12. FCA states (a) and carbon concentration distributions (b) for different values of cin.
Initial carbon concentration directly affects the transformation rate. Ferrite grains
growInitial
fastercarbon concentration
under the condition ofdirectly affects
lower initial the transformation
carbon rate. Ferrite
content in austenite cin . grains
grow faster under the condition of lower initial carbon content in austenite cin.
6. Summary
6. Summary
Phase transformations, especially diffusion phase transformations in carbon steels,
play Phase transformations,
a crucial role in formingespecially diffusion phase
material properties. transformations
Numerical in carbon
modeling remains steels,
pivotal in
play a crucial and
this domain, role in forming
newly material
developed properties.
models employNumerical
diverse modeling remains
approaches pivotal in
and methods to
this domain,
simulate and newlyformation
microstructure developedafter
models employ diverse
transformation. approaches
This paper and
presents themethods
developedto
3D carbonmicrostructure
simulate diffusion model for the simulation
formation of carbon diffusion
after transformation. during
This paper diffusional
presents phase
the devel-
oped 3D carbon diffusion model for the simulation of carbon diffusion during diffusional
phase transformations, which is based on the LBM. A detailed presentation of the devel-
oped carbon diffusion calculation algorithm is provided. Examples of the simulation re-
sults of carbon diffusion and the growth of the new phase in the selected planes of the 3D
modeling space for different simulation conditions are also presented. The simulations
Materials 2024, 17, 674 13 of 14

transformations, which is based on the LBM. A detailed presentation of the developed

carbon diffusion calculation algorithm is provided. Examples of the simulation results
of carbon diffusion and the growth of the new phase in the selected planes of the 3D
modeling space for different simulation conditions are also presented. The simulations
used the D3Q19 LBM scheme and the von Neumann and Moore neighborhoods. The
developed carbon diffusion model incorporates CUDA parallel computing on GPUs. A
fundamental connection was established between the evolved carbon diffusion model and
the microstructure evolution model. The developed carbon diffusion model at the next
stage will also be connected to the developed 3D heat flow model during diffusional phase
transformations. This will be important in the achievement of a comprehensive 3D model
of carbon diffusion and heat flow during diffusional phase transformation.

Author Contributions: Conceptualization, Ł.Ł. and D.S.; Methodology, Ł.Ł. and D.S.; Software,
Ł.Ł.; Validation, Ł.Ł. and D.S.; Formal analysis, Ł.Ł. and D.S.; Investigation, Ł.Ł.; Data curation,
Ł.Ł.; Writing—original draft, Ł.Ł.; Writing—review & editing, D.S.; Visualization, Ł.Ł.; Funding
acquisition, Ł.Ł. All authors have read and agreed to the published version of the manuscript.
Funding: The financial support of the Ministry of Science and Higher Education, Poland (grant
AGH University of Krakow no. 16.16.110.663) and the Polish National Science Centre (grant no.
2016/21/D/ST8/01690) is greatly appreciated.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Data are contained within the article.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Li, W.; Vittorietti, M.; Jongbloed, G.; Sietsma, J. Microstructure–Property Relation and Machine Learning Prediction of Hole
Expansion Capacity of High-Strength Steels. J. Mater. Sci. 2021, 56, 19228–19243. [CrossRef]
2. Dunstan, M.K.; Paramore, J.D.; Fang, Z.Z.; Sun, P. Manipulation of Microstructure and Mechanical Properties during Dehydro-
genation of Hydrogen-Sintered Ti–6Al–4V. Mater. Sci. Eng. A 2019, 764, 138244. [CrossRef]
3. Wen, X.; Wan, M.; Huang, C.; Tan, Y.; Lei, M.; Liang, Y.; Cai, X. Effect of Microstructure on Tensile Properties, Impact Toughness
and Fracture Toughness of TC21 Alloy. Mater. Des. 2019, 180, 107898. [CrossRef]
4. Jagadeesh Babu, S.M.; Kataraki, P.S.; Narayanaswamy, K.S. Phase Transformation and Microstructure Evolution Study in Various
Alloy Systems: An Insight. In IOP Conference Series: Materials Science and Engineering; IOP Publishing: Bristol, UK, 2019; Volume
505. [CrossRef]
5. Fultz, B. Phase Transitions in Materials; Cambridge University Press: Cambridge, UK, 2020. [CrossRef]
6. Porter, D.A.; Easterling, K.E.; Sherif, M.Y. Phase Transformations in Metals and Alloys, 4th ed.; CRC Press: Boca Raton, FL, USA,
2021; pp. 1–556. [CrossRef]
7. Soffa, W.A.; Laughlin, D.E. Diffusional Phase Transformations in the Solid State. In Physical Metallurgy, 5th ed.; Elsevier:
Amsterdam, The Netherlands, 2014; pp. 851–1020.
8. Duhamel, C.; Venkataraman, S.; Scudino, S.; Eckert, J. Diffusionless Transformations. In Basics of Thermodynamics and Phase
Transitions in Complex Intermetallics; World Scientific: Singapore, 2008; pp. 119–145. [CrossRef]
9. Fultz, B. Diffusionless Transformations. In Phase Transitions in Materials; Cambridge University Press: Cambridge, UK, 2014;
pp. 355–382.
10. Dahmen, U. Phase Transformations, Crystallographic Aspects. In Encyclopedia of Physical Science and Technology; Elsevier:
Amsterdam, The Netherlands, 2003; pp. 821–853. [CrossRef]
11. Wang, K.; Yan, C.J.; Yuan, C.H.; Yang, X.C.; Zhao, L.J.; Wang, Q. Progress in Research on Diffusional Phase Transformations of
Fe–C Alloys under High Magnetic Fields. J. Iron Steel Res. Int. 2022, 29, 707–718. [CrossRef]
12. Liu, X.; Li, H.; Zhan, M. A Review on the Modeling and Simulations of Solid-State Diffusional Phase Transformations in Metals
and Alloys. Manuf. Rev. 2018, 5, 10. [CrossRef]
13. Nakajima, K.; Apel, M.; Steinbach, I. The Role of Carbon Diffusion in Ferrite on the Kinetics of Cooperative Growth of Pearlite: A
Multi-Phase Field Study. Acta Mater. 2006, 54, 3665–3672. [CrossRef]
14. Feng, J.; Qingchao, W.; Hongguang, L.; Runzhou, D.; Zhinan, Y.; Chen, C.; Feng, L.; Fucheng, Z. Effect of the Diffused Content of
Carbon on Mechanical Properties of Multilayered Composite Steel. J. Mater. Res. Technol. 2023, 22, 668–682. [CrossRef]
15. Jin, Y.M.; Wang, Y.U.; Khachaturyan, A.G. Macroscopic Energy Barrier and Rate-Independent Hysteresis in Martensitic Transfor-
mations. Acta Mater. 2019, 173, 292–301. [CrossRef]
Materials 2024, 17, 674 14 of 14

16. Campbell, F.C. Elements of Metallurgy and Engineering Alloys; ASM International: Detroit, MI, USA, 2008. [CrossRef]
17. Singh, V.; Patra, S.; Murugan, N.A.; Toncu, D.C.; Tiwari, A. Recent Trends in Computational Tools and Data-Driven Modeling for
Advanced Materials. Mater. Adv. 2022, 3, 4069–4087. [CrossRef]
18. Wolff, M.; Acht, C.; Bochm, M.; Meier, S. Modelling of Carbon Diffusion and Ferritic Phase Transformations in an Unalloyed
Hypoeutectoid Steel. Arch. Mech. 2007, 59, 435–466.
19. Yamanaka, A. Phase-Field Modeling and Simulation of Solid-State Phase Transformations in Steels. ISIJ Int. 2023, 63, 395–406.
[CrossRef]
20. An, D.; Pan, S.; Huang, L.; Dai, T.; Krakauer, B.; Zhu, M. Modeling of Ferrite-Austenite Phase Transformation Using a Cellular
Automaton Model. ISIJ Int. 2014, 54, 422–429. [CrossRef]
21. Lee, S.-J.; Kidder, D.; John, C.; Tyne, V. An Empirical Model for Carbon Diffusion in Austenite Incorporating Alloying Element
Effects. ISIJ Int. 2011, 51, 1903–1911. [CrossRef]
22. Rappaz, M.; Bellet, M.; Deville, M. Numerical Modeling in Materials Science and Engineering; Springer: Berlin/Heidelberg, Germany,
2003; Volume 32. [CrossRef]
23. Robert, C.; Mareau, C. A Comparison between Different Numerical Methods for the Modeling of Polycrystalline Materials with
an Elastic–Viscoplastic Behavior. Comput. Mater. Sci. 2015, 103, 134–144. [CrossRef]
24. Zhu, W.-L.; Yang, S.; Ju, B.-F.; Chen, S.; Xu, Y.; Jiao, Y. A Hybrid Finite-Element and Cellular-Automaton Framework for Modeling
3D Microstructure of Ti–6Al–4V Alloy during Solid–Solid Phase Transformation in Additive Manufacturing. Model. Simul. Mater.
Sci. Eng. 2018, 26, 045011. [CrossRef]
25. Lahiri, A. Phase-Field Modeling of Phase Transformations in Multicomponent Alloys: A Review. J. Indian Inst. Sci. 2022,
102, 39–57. [CrossRef]
26. Lv, S.J.; Wang, S.Z.; Wu, G.L.; Gao, J.H.; Yang, X.S.; Wu, H.H.; Mao, X.P. Application of Phase-Field Modeling in Solid-State Phase
Transformation of Steels. J. Iron Steel Res. Int. 2022, 29, 867–880. [CrossRef]
27. Svyetlichnyy, D. Frontal Cellular Automata for Modelling Microstructure Evolution: Computational Complexity Analysis.
Comput. Mater. Sci. 2023, 230, 112478. [CrossRef]
28. Svyetlichnyy, D.S.; Mikhalyov, A.I. Three-Dimensional Frontal Cellular Automata Model of Microstructure Evolution—Phase
Transformation Module. ISIJ Int. 2014, 54, 1386–1395. [CrossRef]
29. Łach, Ł.; Svyetlichnyy, D. New Platforms Based on Frontal Cellular Automata and Lattice Boltzmann Method for Modeling the
Forming and Additive Manufacturing. Materials 2022, 15, 7844. [CrossRef] [PubMed]
30. Svyetlichnyy, D.; Krzyzanowski, M.; Straka, R.; Lach, L.; Rainforth, W.M. Application of Cellular Automata and Lattice Boltzmann
Methods for Modelling of Additive Layer Manufacturing. Int. J. Numer. Methods Heat Fluid Flow 2018, 28, 31–46. [CrossRef]
31. Łach, Ł. Modeling of Microstructure Evolution during Deformation Processes by Cellular Automata—Boundary Conditions and
Space Reorganization Aspects. Materials 2021, 14, 1377. [CrossRef] [PubMed]
32. Zhu, C.S.; Li, Y.J.; Ma, F.L.; Feng, L.; Lei, P. Three-Dimensional Multi-Phase-Field Simulation of Eutectoid Alloy Based on OpenCL
Parallel. China Foundry 2021, 18, 239–248. [CrossRef]
33. Mohamad, A.A. Lattice Boltzmann Method Fundamentals and Engineering Applications with Computer Codes; Springer: London, UK,
2011; ISBN 978-0-85729-454-8.
34. Rojas, A.C. Analytical Solutions Formulated in the Time Domain for Three-Dimensional Heat Diffusion Equation. Processes 2022,
10, 1472. [CrossRef]
35. Perumal, D.A.; Dass, A.K. A Review on the Development of Lattice Boltzmann Computation of Macro Fluid Flows and Heat
Transfer. Alexandria Eng. J. 2015, 54, 955–971. [CrossRef]
36. Zaitsev, D.A. A Generalized Neighborhood for Cellular Automata. Theor. Comput. Sci. 2017, 666, 21–35. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.