Oxidation of Metals, Vol. 64, Nos.

1/2, August 2005 (© 2005)

DOI: 10.1007/s11085-005-5718-x

Numerical Simulation of Internal Oxidation of Steels

during Annealing Treatments
D. Huin,∗ P. Flauder,∗ and J.-B. Leblond†‡
Received June 24 2004; revised March 25, 2005

This paper presents a new model for simultaneous diffusion and precipitation
of chemical elements in metallic matrices, a scheme for its numerical solution,
and several applications to problems of internal oxidation. The model basically
stands as an extension of the classical Wagner model for internal oxidation of
steels, but is more much general in that it allows for an arbitrary number of diffus-
ing chemical elements, an arbitrary number of precipitate phases with arbitrary
compositions, dependence of diffusion coefficients and solubility products upon
(time-dependent) temperature, etc., thus allowing for a much broader range of
applications. As a counterpart, it is generally impossible to solve the complex,
non-linear equations of the model analytically, but this can be done numerically.
The simple but efficient numerical scheme proposed is based on explicit 1D finite
differences. Experience has shown that this scheme, in spite of its rusticity and
the restrictions it imposes on the time-step, is more efficient than more elabo-
rate strategies based on the finite-element method. The applications presented
are concerned with internal oxidation of steels during annealing processes. The
model and associated numerical scheme allow for evaluation of the amounts of the
various oxide precipitates in the external layer of the sheet. This opens the way,
through numerical parametric studies of the influence of the process parameters
and the chemical composition, to the improvement of existing treatments and the
development of new steel grades.

KEY WORDS: diffusion; precipitation; numerical simulation; internal oxidation; annealing;

numerical optimization.

∗ ARCELOR Research, Auto Products Research Center, Voie Romaine, BP 30320, 57283
Maizières-les-Metz Cedex, France.
† Laboratoire de Modélisation en Mécanique, Université Pierre et Marie Curie (Paris VI),
Tour 65-55, 4 place Jussieu, 75252 Paris Cedex 05, France.
‡ To whom correspondence should be addressed; e-mail: [email protected]
131
0030-770X/05/0800–0131/0 © 2005 Springer Science+Business Media, Inc.
132 Huin, Flauder, and Leblond

INTRODUCTION
Many processes in the metallurgical industry involve simultaneous
diffusion and precipitation of chemical elements: internal oxidation of a
steel sheet during some annealing treatment is just one example. The aim
of such a treatment is to prevent formation of superficial iron oxides or
to eliminate them if they are already present. This is achieved by using a
reducing atmosphere consisting mainly of hydrogen and nitrogen. During
the process, oxygen, traces of which are present in the annealing atmo-
sphere, diffuses inwards, the oxidizable elements diffuse outwards and var-
ious oxide precipitates are formed in the sub-surface alloy region.
The aim of this paper is to present a new model for the simultaneous
diffusion and precipitation of chemical elements in metallic matrices, its
numerical implementation and several applications related to internal oxi-
dation of steel sheets during annealing treatments. Emphasis will be placed
on the last topic. The applications envisaged will notably include a study
of the influence of the oxygen content of the annealing atmosphere upon
the amounts of internal oxides formed, and a systematic investigation, for
a given process, of the types of oxides formed for a wide range of steel
grades.
The first model for internal oxidation was proposed in the seminal
work of Wagner.(1) Although it made several severely restricting hypothe-
ses, such as the presence of a single oxidant, consideration of binary alloys
only, oxidation of the sole most-reactive metal, formation of a single oxide
with a very low solubility product, etc., it already contained, in essence, all
basic ingredients required. Some fundamental features of this model were
its macroscopic character and the assumption of quasi-instantaneous local
thermodynamic equilibrium between the matrix and precipitate phases.
Wagner’s pioneering study gave rise to numerous subsequent works. A
good summary of the theory was given by Rapp.(2) Kirkaldy(3) extended
the model so as to account for non-zero solubility products. Various other
extensions followed; a non-exhaustive list of significant references includes
the works of Whittle et al.,(4) Christ et al.,(5) Gesmundo and Gleeson,(6)
Gesmundo et al.,(7) Huin et al.,(8) , Gesmundo et al.,(9,10) Gesmundo
and Niu(11) and Niu and Gesmundo(12) . The works of Gesmundo and
coworkers, in particular, included extensions to ternary alloys and for-
mation of two oxides (Niu and Gesmundo;(12) the same problem was
considered by Huin et al.(8) ), and also to the presence of two oxidants
(Gesmundo and Niu(11) ). A very comprehensive critical discussion of the
limitations of Wagner’s model and its improved variants was provided by
Douglass.(13)
Numerical Simulation of Internal Oxidation of Steels 133

These works generally used analytical methods to solve the equa-

tions of the model. Although numerical solutions have been provided for
instance by Bongartz et al.,(14) Christ et al.,(5) Rank and Weinert(15) and
Fortunier et al.,(16) it seems that use of numerical methods to solve this
type of problems has remained comparatively marginal. In contrast, the
focus of this work is on numerical methods and their applications. Indeed
it is clear that they offer the only possible means of studying complex sit-
uations (with two or more oxidizable elements presence of “mixed” oxides
involving several such elements each, variable temperature, etc.), such as
encountered in industrial practice.
The model used is presented in Section “Theoretical Model”. It
belongs to the general category of models for diffusion with chemical
reactions defined in Crank’s(17) book. It basically stands as an extension
of Wagner’s pioneering model, and in particular retains the macroscopic
character of the approach and the assumption of local thermodynamic
equilibrium. However it is presented in a quite general format (arbitrary
numbers of oxidant and oxidizable elements, arbitrary number of oxidized
phases with arbitrary compositions, dependence of diffusion coefficients
and solubility products upon time-varying temperature, etc.), thus allowing
for a much greater variety of applications. It is worth noting for instance
that although applications presented here are all concerned with internal
oxidation during annealing processes, the model could equally well be used
to study carburization or nitridation.
One inevitable counterpart of the generality of the model is the
impossibility of solving its equations analytically except in very special
cases. However Section “Numerical Implementation” proposes a simple
method to do so numerically, based on elementary, explicit 1D finite
differences. This scheme is admittedly crude and places restrictions on
the time-step due to problems of numerical stability. However, experience
(Fortunier et al.,(16) ) has shown that the more elaborate, finite element
method breaks down, resulting in absence of convergence of the iterative
procedure used to deal with non-linearities, when precipitation becomes
very important, as is the case in many applications.
Applications are presented in Section “Applications”. We first deter-
mine, in subsection “Preliminary Remark”, under which √ conditions the
solution happens to depend on the sole variable x/ t. Subsection “Iso-
thermal Annealing—Precipitation of Simple Oxides” considers three exam-
ples of such simple cases, and compares numerical results to the relevant
analytic or semi-analytic solutions provided by Wagner(1) and Huin et
al.,(8) in order to assess the validity of the numerical scheme. More com-
plex cases not amenable to analytic solutions are considered next. Subsec-
tion “Anisothermal Annealing—Precipitation of Simple Oxides” envisages
134 Huin, Flauder, and Leblond

that of anisothermal annealing of some ternary Fe–Mn–Si alloy, account

being taken only of the “simple” oxides (containing only one oxidizable
element each) MnO and SiO2 . Precipitation of the extra, “mixed” oxide
Mn2 SiO4 is considered in Subsection “Anisothermal Annealing—Precipi-
tation of Simple and Mixed Oxides”, with a comparison between numer-
ical and experimental results. Subsection “Influence of the Dew Point of
the Annealing Atmosphere” illustrates the influence of the oxygen content
of the annealing atmosphere. Subsection “Influence of Extra Alloying Ele-
ments—Example of Aluminium” exemplifies the influence of extra alloy-
ing elements, aluminum serving as a typical example. Finally Subsection
“Effect Mn/Si Ratio” presents a systematic investigation of internal pre-
cipitation of oxides for a given annealing treatment but for a wide range
of steel grades.

THEORETICAL MODEL
General Hypotheses and Notations
The material consists of several phases (Fig. 1). The first one con-
sists of a “nob1e” metal, Fe generally, plus small amounts of alloying ele-
ments: Mn, Si,O. . . (hypothesis of dilute solution). The noble metal does
not combine with other chemical elements, so that its role only consists of
providing a “matrix” phase in which diffusion and precipitation of other
elements take place. There are also small amounts of various precipitate
phases composed only of alloying elements (MnO, SiO2 , . . . ).
The model is macroscopic in character; the “elementary volumes”
it considers are sufficiently large to contain the matrix phase plus all
precipitate phases (Fig. 1). Transfers of chemical elements between the
matrix and precipitate phases are accounted for by assuming local ther-
modynamic equilibrium in each elementary volume, without accounting

Fig. 1. Schematic “elementary volume” considered

by the model.
Numerical Simulation of Internal Oxidation of Steels 135

for the local variation of chemical composition of the matrix near each
precipitate.
The number of alloying elements is noted ne , and these elements are
denoted with Latin indices i, j . The atomic mass of element i is denoted
Mi . Its total mass fraction (in all its possible forms, that is dissolved in
the matrix or embedded in precipitates) is denoted Fi . Its concentration in
the matrix (dissolved mass fraction) is denoted Ci . Diffusion of elements
is assumed to take place only within the matrix phase; the diffusion coeffi-
cient of element i is denoted Di , and is an Arrhenius-type function of
temperature.
The number of possible precipitate phases is noted np , and these
precipitates are denoted with a Greek index α. The molecular mass of
precipitate α is denoted Mα . Its mass fraction is denoted Pα . Its solubil-
ity product is denoted Kα and is an Arrhenius-type function of tempera-
ture. The stoechiometric coefficient of element i in precipitate α (number
of atoms of this element in each molecule of this precipitate) is denoted
Niα .
Counting atoms of element i in all their possible forms (dissolved in
the matrix or embedded in precipitates) in an elementary volume, one gets
the balance equation

np
Fi Ci
Pα
= + Niα . (1)
M i Mi Mα
α=1

Diffusion Equations
The flux Ji of element i in the matrix is assumed to be given by the
familiar expression

Ji = −Di gradCi . (2)

This formula implies equating the activity of an element to its concentra-

tion; the underlying approximations have been discussed many times and
are reasonable for dilute solutions as considered here. The conservation
law for the total number of atoms of element i (which is unaffected by
chemical reactions) in an arbitrary volume V then reads
  
∂
Fi dV = − Ji · ndS = Di gradCi · ndS
∂t V  ∂V ∂V

= div(Di gradCi )dV ,

V
136 Huin, Flauder, and Leblond

where n denotes the unit normal outward vector to the boundary ∂V . The
diffusion equations follow from the local form of this law:
∂Fi
= div(Di gradCi ). (3)
∂t
The influence of the precipitation rates is implicit in this equation, but it
can be made explicit by using the balance law (1) to transform the left-
hand side; one thus gets
np
∂Ci
Niα ∂Pα
= div(Di grad Ci ) − Mi . (4)
∂t Mα ∂t
α=1

The sum in the right-hand side here acts as a “source” or “sink” term due
to chemical reactions.

Thermodynamic Equilibrium
Detailed mechanisms of precipitation or re-dissolution of precipitates
are disregarded. Instantaneous local thermodynamic equilibrium between all
phases within each elementary volume, governed by the laws of mass
action, is assumed instead. Thus for every α,

 
ne

 Pα > 0 and CiNiα = Kα

 i=1
or (5)

 
ne

 N
 Pα = 0 and C i ≤ Kα .
iα

i=1

Boundary Conditions
For each element i, there are two possible basic boundary conditions :
prescr
Ci = Ci on∂ (6)

or
prescr
Ji · n = −Ji on∂, (7)
prescr
where ∂ denotes the boundary of the domain  considered and Ci
prescr
and Ji the prescribed concentration and the prescribed input flux.
Boundary conditions can also be of “mixed type”, i.e. Eq. (6) on some
part of ∂, (7) on the complementary part.
Numerical Simulation of Internal Oxidation of Steels 137

NUMERICAL IMPLEMENTATION
Early numerical applications of the model (Fortunier et al.(16) ) based
on the finite element method encountered severe difficulties in cases involv-
ing important precipitation, i.e. small solubility products. Autesserre,(18)
Huin and Lehmann(19) and Flauder and Huin(20) later developed a much
simpler scheme based on explicit 1D finite differences which revealed much
better adapted to this type of problems, in spite of its rusticity and the
inevitable ensuing restrictions on the time-step.
We thus consider only 1D problems, the spatial coordinate being
denoted x. The temperature is considered as independent of x (uni-
form), though dependent on t. This approximation is reasonable for the
applications described below, because although the industrial processes
considered will be anisothermal, the sub-surface region where diffusion
and precipitation take place is so shallow that the temperature is nearly
uniform within it at each instant. Thus the Di are uniform and the diffu-
sion equations (3) reduce to

∂Fi ∂ 2 Ci
= Di . (8)
∂t ∂x 2

Temporal and Spatial Discretization of Diffusion Equations

All quantities at time t (denoted with an upper index(0) ) being
assumed to be known, the problem is to determine all quantities at time
t + t (denoted without any index).
We introduce some spatial discretization involving N nodes x1 =
0, x2 , . . . , xN (Fig. 2). In practice, node 1 will represent the outer surface
of the sheet in contact with the annealing atmosphere, and node N will be
located deep in its thickness. The distance xk = xk+1 − xk between consec-
utive nodes need not be identical. (In practice the mesh is refined near the
outer surface).

Fig. 2. Spatial discretization.

138 Huin, Flauder, and Leblond

Equation (8), discretized in time using an explicit Euler scheme of

order 1 and in space using finite differences, read
(0)
 (0) (0)
Fi (xk ) − Fi (xk ) 2Di Ci (xk+1 ) − Ci (xk )
=
t xk−1 + xk xk
(0) (0)

Ci (xk ) − Ci (xk−1 )
− . (9)
xk−1

Note that these equations do not apply at the boundary nodes x1 and xN .
Since the algorithm is explicit, time-steps smaller than the stability
1
limit must be used. In practice, t is taken as 10 of the smallest of the
2
quantities (xk ) /Di .

Calculation of Thermodynamic Equilibrium

The Fi being known at nodes x2 , x3 , . . . , xN−1 by Eq. (9), one can
calculate the Ci and Pα at the same points by solving Eq. (5), using the
following procedure. Assume first that the set A of indices α correspond-
ing to existing precipitates (Pα > 0) is known; let p (≤ np ) denote the num-
ber of indices in this set. Then the equalities ni=1 e
CiNiα = Kα , written for
all α ∈ A, provide p equations on the p unknowns Pα , α ∈ A, the Ci being
expressed in terms of the Fi and these non-zero Pα using Eq. (1). This set
of equations can be solved numerically using Newton’s method.
The problem is thus reduced to determining the set A of indices of
existing precipitates. This can be done using an iterative procedure. Indeed,
once the Pα , α ∈ A, and the Ci have been found for a given A (i.e., when
the presumed state of thermodynamic equilibrium has been found for a
given hypothesis on existing precipitates), one must check (i) that all sup-
posedly existing precipitates (α ∈ A) have a non-negative Pα and (ii) that
all supposedly absent precipitates (α ∈ / A) respect the inequality ni=1
e
CiNiα ≤
Kα . If condition (i) is violated, those supposedly existing precipitates hav-
ing Pα < 0 must be removed from A for the next iteration. If condition
(ii) is violated, those supposedly absent precipitates having ni=1 e
CiNiα > Kα
must be added to it.
It must be stressed that although existence and uniqueness of the state
of thermodynamic equilibrium is guaranteed from a purely mathematical
point of view (Leblond(21) ), actual determination of this state is often a
very difficult numerical problem. The major difficulty arises from often
exceedingly small values of the solubility products at low temperatures,
which generate various numerical drawbacks (absence of convergence of
Newton’s method, singular tangent matrix,. . . ). Various numerical tricks
Numerical Simulation of Internal Oxidation of Steels 139

must be used to solve these problems. Also, a lot of CPU time being spent
in the routine evaluating thermodynamic equilibrium, it is necessary to
optimize the choice of the first trial hypothesis on existing precipitates, so
as to minimize the number of necessary iterations.

Treatment of Boundary Conditions

Boundary conditions are different for oxygen (element 1 in this Sub-
section and the next one) and oxidizable elements (elements 2, 3, . . . , ne in
this Subsection and the next one), and also at nodes x1 and xN . At node
xN , deep in the steel sheet, it is assumed that nothing occurs during the
whole treatment so that the boundary conditions read

C1 (xN ) = 0; Ci (xN ) = Cicore , i = 2, 3, . . . , ne , (10)

where the Cicore denote the core concentrations of oxidizable elements

(nominal concentrations defined by the grade of the steel). The Pα (xN ) are
zero and the Fi (xN ) are identical to the Ci (xN ).
At node x1 , it is assumed that the concentration of oxygen is pre-
scribed by its partial pressure in the annealing atmosphere (quick chemical
equilibrium) and that the fluxes of oxidizable elements are zero:
∂Ci
C1 (x1 ) = C1surf ; Di (x1 ) = 0, i = 2, 3, . . . , ne (11)
∂x
where C1surf is the surface concentration of oxygen defined by its solubility
in iron. In order to discretize Eq. (11)2 , it seems natural to enforce iden-
tical Ci -values (i = 2, 3, . . . , ne ) at nodes x1 and x2 . However numerical
experience shows that the concentration profiles are then rather nasty, with
a markedly discontinuous slope at node x2 . We therefore use an equation
leading to nicer concentration profiles, derived from the following “physi-
cal” reasoning. Consider the zone between the outer surface of the sheet
and the midpoint 21 (x1 + x2 ) of the segment (x1 , x2 ) (Fig. 2). The input flux
per unit time of element i (i = 2, 3, . . . , ne ) through the right boundary of
(0) (0)
this zone is Di [Ci (x2 ) − Ci (x1 )]/x1 , and there is no flux through the
left boundary. Since the length of the zone is 21 x1 , the discretized time-
derivative of Fi at node x1 follows:

2Di
(0) 
(0)
Fi (x1 ) − Fi (x1 ) (0)
= C i (x 2 ) − C i (x 1 ) , i = 2, 3, . . . , ne . (12)
t (x1 )2
This equation yields the Fi (x1 ) (i = 2, 3, . . . , ne ). The Ci (x1 ) (i = 2, 3, . . . , ne )
and Pα (x1 ) then follow from calculation of the state of thermodynamic
140 Huin, Flauder, and Leblond

equilibrium. The fact that it is C1 that is known at node x1 , instead of

F1 , can be dealt with by moving C1 from the left-hand side to the right-
hand side in Eq. (5); these equations are then solved by the same routine
as usual, the number of elements being set to ne − 1 and “modified” solu-
bility products being used. Finally F1 (x1 ) is deduced from Eq. (1) for i = 1.

Summary of Numerical Procedure Between Instants t and t+t

• At nodes x2 , x3 , . . . , xN−1 , Eq. (9) yields the Fi (xk ). The Pα (xk ) and
Ci (xk ) then follow by solving Eqs. (5) as sketched in Subsection
“Calculation of Thermodynamic Equilibrium”.
• At node xN , Eq. (10) provides the Ci (xN ). The Pα (xN ) are zero and
the Fi (xN ) are identical to the Ci (xN ).
• At node x1 , Eq. (12) yields the Fi (x1 )(i = 2, 3, . . . , ne ). The Pα (x1 )
and Ci (x1 ) (i = 2, 3, . . . , ne ) then follow by solving Eqs. (5) as
sketched in Subsection “Treatment of Boundary Conditions”,
C1 (x1 ) being known by Eq. (11)1 . Equation (1) with i = 1 finally
provides F1 (x1 ).

APPLICATIONS
In all simulations presented below, the steel sheet will be in contact
with an annealing atmosphere with constant dew point (DP). The defini-
tion of this quantity and the calculation of the dissolved concentration of
oxygen imposed on the surface, C1surf ≡ CO surf , from the DP and the cur-

rent temperature T, are explained in Appendix A. In most simulations,

the annealing process will be anisothermal so that CO surf will vary in time

according to the relations given in Appendix A, even though the DP will

be constant. Only in those simulations corresponding to isothermal pro-
cesses will COsurf be constant in time.

All diffusion and precipitation data used in the calculations are pro-
vided in Appendix B.

Preliminary Remark
Let us first determine in which circumstances the solution happens √ to
depend on the two variables x, t only through the single quantity x/ t ≡ u.
If such is the case, Eq. (8) becomes (u/2)dFi /du + Di d 2 Ci /du2 = 0, where
dFi /du and d 2 Ci /du2 are functions of u only. This is possible only if the Di
are also functions of the sole variable u, and this occurs only if the temper-
ature is constant. If such is the case, Eqs. (1) and (5) are compatible with
Numerical Simulation of Internal Oxidation of Steels 141

the assumed dependence of the solution upon the sole variable u since the
Mi , Mα , Niα and Kα do not vary in time.
Consider now the initial and boundary conditions. Initial conditions
are prescribed at all points (x > 0, t = 0+ ). All such points have u = +∞.
Therefore dependence of the solution upon the sole variable u requires
that initial conditions be independent of x. This condition is satisfied since
the initial total and dissolved fractions of oxygen and the initial frac-
tions of oxide precipitates are zero, and the initial total and dissolved frac-
tions of oxidizable elements are equal to the nominal concentrations Cicore .
Boundary conditions are prescribed at all points (x = 0, t > 0) and (x =
+∞, t > 0). The former points all have u = 0; therefore dependence of the
solution upon the sole variable u requires that boundary conditions on the
surface be independent of t, and this is true only if the dissolved fraction
of oxygen imposed, Cosurf , is independent of t. The latter points all have
u = +∞; therefore dependence of the solution upon the sole variable u also
requires that boundary conditions deep in the sheet be independent of t
and identical to initial conditions. This is true for all elements.
Two conditions must thus be √ met for the solution of the problem to
depend upon the sole variable x/ t : (i) the annealing process must be
isothermal; and (ii) the dissolved concentration of oxygen imposed on the
surface must not vary in time. In the cases considered here, fulfilment of
the first condition automatically entails fulfilment of the second one, since
the annealing atmospheres considered have a constant DP.

Isothermal Annealing—Precipitation of Simple Oxides

We first give results for simulations of isothermal
√ processes, for which
the solution depends upon the sole variable x/ t. The aim here is to
assess the validity of the numerical procedure through comparison of
numerical results with Wagner’s(1) famous analytic solution and Huis et
al.’s(8) semi-analytic solution. The former solution, which is summarized
for completeness in Appendix C, is for a single oxidant, a single oxidiz-
able element and a single precipitate with a very low solubility product.
The latter solution, which is presented in Appendices D and E, extends
upon the former by considering non-zero solubility products and several
oxidizable elements. It was developed only in the case of “simple” oxides
(i.e., involving a single oxidizable element each), but it could be extended
as well to “mixed” oxides (i.e., involving several oxidizable elements each),
the only necessary condition being that the sequence, in time or space, of
successive combinations of oxides must be known a priori.
We first consider the case of isothermal annealing of a 0.12% Al
(in weight) steel, involving diffusion of O and Al and precipitation of
142 Huin, Flauder, and Leblond

Al2 O3 . The DP of the annealing atmosphere is −40◦ C and the temper-

ature is 800◦ C. The corresponding values of the diffusion coefficients,
the solubility product and the superficial concentration of oxygen are
Do = 8.01 (µm)2 s−1 , DAl = 5.56 10−4 (µm)2 s−1 , KAl2 O3 = 2.11 × 10−18 (ppm)5
and Cosurf = 9.23 × 10−3 ppm. The value of KAl2 O3 is sufficiently small for
Wagner’s solution to be applicable. The distribution of the concentration
of dissolved oxygen is not shown because it is trivial; in both the numer-
ical simulation and the analytical solution, this concentration decreases
almost linearly from the surface to the oxidation front, and is zero deeper
in the sheet. Figure 3 shows the distribution of the total mass fraction of
Al at time t = 60 s. The agreement between the numerical results and the
Wagner solution is excellent. This shows that suitable numerical methods
can efficiently deal with internal oxidation problems even in difficult cases
where the chemical reaction is confined to some small region of space,
resulting in quasi-discontinuities of some of the unknowns.
We now consider isothermal annealing of a 1.235% Mn steel, involv-
ing diffusion of O and Mn and precipitation of MnO. The DP of the
atmosphere is −40◦ C and the temperature is 800◦ C. The values of the
diffusion coefficients, the solubility product and the superficial concentra-
tion of Oxygen are then Do = 8.01(µm)2 s−1 , DMn = 9.06l0−4 (µm)2 s−1 ,
KMnO = 7.19 (ppm)2 and Cosurf = 9.23 × 10−3 ppm. Wagner’s solution is not
applicable because the solubility product is too large, but that of Huin
et al. expounded in Appendix D is. Figures 4 and 5 show the distributions
of the dissolved mass fraction of O and the total mass fraction of Mn
at time t = 60 s. Again, the agreement between the theoretical and numer-
ical solutions is excellent. From a more physical point of view, the main

Fig. 3. Isothermal treatment, precipitation of Al2 O3 —total frac-

tion of Al.
Numerical Simulation of Internal Oxidation of Steels 143

Fig. 4. Isothermal treatment, precipitation of MnO—fraction of

dissolved O.

Fig. 5. Isothermal treatment, precipitation of MnO—total fraction

of Mn.

difference with respect to the classical Wagner solution lies in the fact
that the total fraction of Mn is no longer constant in the oxidized zone,
because the fraction of MnO is itself no longer constant. This effect is inti-
mately tied to the non-zero value of the solubility product and is explained
in detail in Appendix D.
We finally consider isothermal annealing of a 1.225% Mn – 0.075%
Si steel, involving diffusion of O, Mn and Si and precipitation of the
sole simple oxides MnO and SiO2 . The DP of the atmosphere is −40◦ C
and the temperature is 800◦ C. The values of the diffusion coefficients,
144 Huin, Flauder, and Leblond

Fig. 6. Isothermal treatment, precipitation of MnO and SiO2 —total

fractions of Mn and Si.

the solubility products and the superficial concentration of oxygen are

DO = 8.01(µ m)2 s−1 , DMn = 9.06 10−4 (µm)2 s−1 , DSi = 1.48 10−3 (µm)2 s−1 ,
KMnO = 7.19 (ppm)2 , KSiO2 = 3.07 10−5 (ppm)3 and Cosurf = 9.23 10−3 ppm.
Huin et al.’s solution presented in Appendix E is applicable to this case.
Figure 6 shows the total mass fractions of Mn and Si at time t = 90 s. The
agreement between theoretical and numerical results is again very good.
(The theoretical solution was itself successfully compared to experiments
by Loison et al.(22) ). Note also that whereas enrichment in Si occurs over
a non-negligible distance, enrichment in Mn occurs on the sole surface
node (and is therefore invisible in the figure). This is because precipitation
of MnO occurs over a much smaller distance than precipitation of SiO2 ,
owing to the rather large solubility product of MnO.

Anisothermal Annealing—Precipitation of Simple Oxides

From now on, we shall consider only anisothermal processes. No ana-
lytic or semi-analytic solutions are available for such cases. The thermal
history will always be the same and correspond to some typical industrial
treatment: the temperature will increase linearly from 20◦ C at time t = 0
to 800◦ C at time t = 130 s, remain constant at 800◦ C from t = 130 s to
170 s, decrease back linearly from 800◦ C at t = 170 s to 500◦ C at t = 270 s,
and finally remain constant at 500◦ C from t = 270 s to t = 300 s.
In this Subsection, we consider anisothermal annealing of a 1.235%
Mn – 0.12% Si steel, involving diffusion of O, Mn and Si and precipitation
of the simple oxides MnO and SiO2 . The DP of the atmosphere is –40 ◦ C.
Numerical Simulation of Internal Oxidation of Steels 145

Fig. 7. Anisothermal treatment, precipitation of simple oxides—total fraction

of Mn.

Figure 7 shows the total fraction of Mn at the end of heating, the end
of the isothermal stage and the end of the treatment. (The Si profile, not
shown, is qualitatively similar).
√ The fact that the solution does not depend
upon the sole variable x/ t is clearly apparent; indeed the curves cannot
be deduced from each other through contraction or dilation of the space
coordinate since they have different minima. Enrichment in Mn near the
surface and deeper depletion are already apparent at the end of heating
and even more so at the end of the isothermal period. The depletion devel-
ops a little further during final cooling.

Anisothermal Annealing—Precipitation of Simple and Mixed Oxides

We consider a 1.225% Mn–0.132% Si steel subject to the same aniso-
thermal treatment as before but now account for the possible formation of
the additional, “mixed” oxide Mn2 SiO4 . The calculation evidences forma-
tion of MnO, plus Mn2 SiO4 rather than SiO2 . Figure 8 shows the total
fractions of Mn and Si at the end of the treatment. Again, deep depletion
in both Mn and Si and enrichment near the surface can be observed; the
dimensions of the enriched and depleted zones are almost the same for the
two elements.
Experiments were also performed for this specific steel grade and
annealing treatment; the final profiles of Mn and Si were measured as
functions of position by secondary-ion mass spectroscopy. The total frac-
tion of Mn was found to be minimum at a distance of 0.08 µm from
146 Huin, Flauder, and Leblond

Fig. 8. Precipitation of simple and mixed oxides—total fractions of Mn and

Si.

the sheet surface, in acceptable agreement with the numerical simulation.

However the minimum observed experimentally was less marked than in
the simulation, the ratio of the minimum fraction of Mn over its nominal
value being of the order of 1/3 versus 1/8 in the simulation. Accurate
measurement of the total fraction of Si was difficult because of its scarce-
ness, but very significant enrichment in Si near the surface could be
observed, in agreement with the numerical results.

Influence of the Dew Point of the Annealing Atmosphere

We now illustrate the use of the model and the associated numeri-
cal scheme as a predictive tool through simulations of imaginary annealing
treatments and/or steel grades.
It was first shown by Wagner,(1) in the simple case of a single oxidant,
a single oxidizable element and a single oxide with a very low solubility
product, that the consequences of an increase of the concentration of oxi-
dant on the external surface may be non-trivial: indeed it may result in a
decrease, contrary to intuition, of the amount of oxides formed. It is there-
fore difficult to anticipate what the effect of an increase of the DP of the
annealing atmosphere will be, and numerical simulations are needed to set-
tle this question.
We consider the same steel and thermal history as in
Subsection “Anisothermal Annealing—Precipitation of Simple Oxides”,
but now account for possible precipitation of Mn2 SiO4 and consider two
atmospheres with DP of −40◦ C and +15◦ C. Figures 9 and 10 show the
Numerical Simulation of Internal Oxidation of Steels 147

Fig. 9. Annealing treatments with DP of −40◦ C and +15◦ C—total fraction

of Mn.

Fig. 10. Annealing treatments with DP of −40◦ C and +15◦ C—total fraction
of Si.

final total fractions of Mn and Si. Enrichment in both of these elements

near the surface is less important for the treatment with a high DP .
Figure 11 shows the fractions of Mn for the two treatments. The
bottom curve represents the fraction of dissolved Mn atoms, the middle
148 Huin, Flauder, and Leblond

Fig. 11. Annealing treatments with DP of −40◦ C and +15◦ C—fraction of Mn atoms.

one that of atoms dissolved or trapped in Mn2 SiO4 precipitates, and the
top one that of atoms dissolved or trapped in Mn2 SiO4 or MnO pre-
cipitates. The vertical distance between the first two curves thus repre-
sents the fraction of Mn atoms trapped in Mn2 SiO4 precipitates, and
that between the last two curves the fraction of atoms trapped in
MnO precipitates. Changing the DP from −40◦ C to +15◦ C results
in a decrease by a factor of 10 of the amount of atoms trapped in
oxides.

Influence of Extra Alloying Elements—example of Aluminum

In order to illustrate the influence of extra elements, we consider
the same grade as in Subsections “Anisothermal Annealing—Precipitation
of Simple Oxides” and “Influence of the Dew point of the Annealing
Atmosphere” but now account for the presence of 0.06% or 0.12% of
Al. We repeat the simulation of Subsection “Influence of the Dew point
of the Annealing Atmosphere” for the treatment with a DP of +15◦ C,
now including diffusion of Al and possible precipitation of Al2 O3 and
MnAl2 O4 . (There are thus 5 a priori possible precipitates, MnO, SiO2 ,
Mn2 SiO4 , Al2 O3 and MnAl2 O4 , but the phase rule in fact prohibits actual
formation, of more than 3 of these). Figures 12 and 13 show the final total
Numerical Simulation of Internal Oxidation of Steels 149

Fig. 12. Comparison of Mn profiles obtained without and with Al.

Fig. 13. Comparison of Si profiles obtained without and with Al.

fractions of Mn and Si; results obtained without Al are also shown for ref-
erence. The presence of Al does not change the qualitative shape of the
Mn and Si profiles but slightly enhances the enrichment in these elements
near the surface while reducing the depth of the enriched zone.
150 Huin, Flauder, and Leblond

Fig. 14. (Dissolved Si)—(dissolved Mn) diagram —T = 800◦ C, DP = −40◦ C.

Effect of Mn/Si Ratio

We finally present, for a given annealing treatment, a systematic study
of the types of internal oxides formed for Mn–Si steels with a wide range
of chemical compositions. The aim here is to provide a tool useful for the
development of new steel grades. Results will be presented in the form of
diffusion paths in a (dissolved Si)–(dissolved Mn) diagram. This diagram
differs slightly from the usual one since it does not include oxygen; some
explanatory comments are probably useful here.
Figure 14 shows such a diagram, at a temperature of 800◦ C and
for an annealing atmosphere with a DP of −40◦ C. Let P denote the
current point in this diagram. Suppose x = 0 (CO = CO surf . The line

ABCD represents, as usual, the possible positions of P if precipitation

occurs. (Segments AB, BC and CD correspond to precipitation of SiO2 ,
Mn2 SiO4 and MnO respectively, and points B and C to coexistence of
SiO2 and Mn2 SiO4 , and Mn2 SiO4 and MnO respectively). Now suppose
x > 0 (CO < CO surf ). The line ABCD moves upwards and to the right and

becomes A B  C  D  . The loci of points B and C , when x varies, are

lines BE and CF . These lines thus represent possible locations of P if
SiO2 and Mn2 SiO4 , or Mn2 SiO4 and MnO coexist, irrespective of the
position of x.
Now consider the diffusion path for an isothermal treatment with
constant DP . It was shown in Subsection “Preliminary Remark” that
Numerical Simulation of Internal Oxidation of Steels 151

Fig. 15. Diffusion paths for various grades in the (dissolved Si) —(dissolved Mn)
diagram.

√
for such a treatment, the solution depends on the sole variable x/ t so
that it is equivalent to consider variations of t for fixed x or variations
of x for fixed t; let us consider the first possibility. At t = 0, there is
no oxygen, line A B  C  D  is infinitely remote upwards and to the right,
all oxides are absent, and CSi = CSi core , C
Mn = CMn . For t > 0, CO starts
core
   
increasing, so that line A B C D starts moving downwards and to the left.
Point P moves even though precipitation of oxides does not yet occur
at the position considered, because precipitation near the surface induces
diffusion of oxidizable elements everywhere. For some t, A B  C  D  hits P
and precipitation starts at the position considered; as t further increases,
A B  C  D  gets closer and closer to ABCD, sweeping P in its motion. In
the limit t → +∞, A B  C  D  ultimately coincides with ABCD; P lies some-
where on this line, and its position determines which oxides are finally
present.
It is important to note that the initial position of point P (defined by
the grade of the steel) with respect to lines BE and CF does not allow to
predict which oxides will ultimately be present, because nothing prevents
this point from crossing these lines.
Figure 15 shows the diffusion paths obtained numerically for various
grades for the anisothermal treatment of Subsection “Anisothermal Anneal-
ing — Precipitation of Simple Oxides”. These paths are plotted only up to
the end of the isothermal stage, in the (dissolved Si)–(dissolved Mn) dia-
152 Huin, Flauder, and Leblond

gram corresponding to the temperature of 800◦ C imposed then.1 Thin lines

represent paths and thick ones limits between domains of composition cor-
responding to final formation of specific combinations of oxides. (This is
emphasized by switching the symbols representing paths from full lines to
hatched ones and vice versa from one domain to the next one). Thus, if
core , C core ) of the path lies in rectangle AGHB, its end-
the starting point (CSi Mn
point lies on segment AB so that the sole oxide SiO2 is present at the end
core , C core ) lies in region BHI, the path ends at point
of the treatment; if (CSi Mn
B and both SiO2 and Mn2 SiO4 are finally present; if it lies in BIJC, only
Mn2 SiO4 is finally present; if it lies in CJK, both Mn2 SiO4 and MnO are
finally present; and if it lies in rectangle CKLD, only MnO is finally present.
Lines BE and CF of Figure 14 have also been plotted in Fig. 15.
The domains defined by these lines, instead of lines BI and CJ , are quite
different from those actually computed. This confirms the remark made
above that the position of point (CSi core , C core ) with respect to lines BE and
Mn
CF cannot be used to deduce which oxides will finally be present.

SUMMARY AND PERSPECTIVES

A phenomenological model for the simultaneous diffusion and pre-
cipitation of chemical elements in metals has been described. This model
is an extension of Wagner1 ’s classical model for internal oxidation, allow-
ing for a much broader range of applications. An efficient associated
numerical scheme based on 1D explicit finite differences was proposed.
Various applications related to internal oxidation of a steel sheet
during some annealing treatment were presented. The applicability of the
model to complex practical situations including formation of mixed oxides
and isothermal processes was first demonstrated for low-alloy Mn–Si
steels. Applications to imaginary treatments and/or steel grades, evidencing
the ability of the approach to optimize existing treatments and help in the
development of new grades, were then presented. We first studied the effect
of an increase of the oxygen content of the annealing atmosphere. The
influence of extra alloying elements was illustrated next, aluminum serv-
ing as a typical example. The types of internal oxides formed were finally
determined for a given treatment but for a wide range of steel grades. This
allowed for the definition of domains of chemical composition leading to
formation of specific combinations of internal oxides.
This work is liable to various extensions. Three improvements of the
model are currently under progress. The first one consists of accounting

1 There is an approximation here since the temperature is not constant during the √
treatment,
but simulations show that the solution depends almost solely on the variable x/ t in this
case.
Numerical Simulation of Internal Oxidation of Steels 153

for hindering of diffusion by precipitates by considering the diffusion

coefficients as decreasing functions of their volume fractions. The second
one consists of accounting for simultaneous external oxidation by incor-
porating the formation of a superficial layer of FeO, the outward diffusion
of iron through this layer, and its oxidation on the external surface, into
the model. The third one consists of relaxing the hypothesis of instanta-
neous local thermodynamic equilibrium by using some heuristic equation
for the size of the precipitates and incorporating an influence of this size
into some evolution equations for their volume fractions.

ACKNOWLEDGMENT
The authors express their sincere thanks to Ph. Autesserre, who was a
great help in the early stages of development of the numerical procedure,
and to F. Mudry and P.V. Riboud for their moral and financial support.

APPENDIX A: RELATION BETWEEN THE DEW POINT OF THE

ANNEALING ATMOSPHERE AND THE SUPERFICIAL
CONCENTRATION OF OXYGEN
The dew point of the annealing atmosphere is the temperature at
which the first droplet of liquid water or nucleus of ice appears, when this
atmosphere is cooled down under conditions of constant partial pressure
of water vapor. It thus depends on the water content of the atmosphere,
and in fact serves as a conventional measure of this content. The water
content in turn determines the oxygen content of the atmosphere at any
given temperature, through chemical equilibrium between water vapor and
gaseous oxygen and hydrogen. The oxygen content finally determines the
superficial concentration of dissolved oxygen on the steel sheet, through
the solubility of oxygen in α iron.
The calculation of this superficial concentration follows these steps in
a natural way. First, the following formulae, which were adjusted using
experimental data provided by the Thermodata(23) bank, provide the satu-
ration vapor pressure of water (psatH2 O ), in atm, at the temperature of the
dew point (DP ), in ◦ C:

9.80 DP /(273.8 + DP ) − 2.22 if DP ≤ 0◦ C,
log10 psat H2 O = (A.1)
7.58 DP /(240 + DP ) − 2.22 if DP > 0◦ C.

This is the partial pressure of water vapor in the annealing atmosphere,

when it is prepared at the temperature DP . The following formula (Rist et
al.(24) ) then gives the partial pressure of oxygen (pO2 ) of this atmosphere,
154 Huin, Flauder, and Leblond

in atm, at the temperature (T ) of the annealing process, in K:


1 13088 psatH2 O
log10 pO2 = 3.00 − + log10 . (A.2)
2 T pH2
In this equation pH2 denotes the partial pressure of hydrogen. The anneal-
ing atmospheres considered in this paper consist of 95% nitrogen and 5%
hydrogen (in moles) (plus small amounts of oxygen and water vapor),
and their total pressure is 1 atm, so that pH2 = 0.05 atm.2 The follow-
ing formula, taken from Swisher and Turkdogan25 (with a slight simpli-
fication) finally provides the superficial mass concentration of dissolved
oxygen on the steel sheet (COsurf ), in ppm:

surf 9398 1
log10 CO = 1.00 + + log10 pO2 . (A.3)
T 2

APPENDIX B: DIFFUSION AND PRECIPITATION DATA

Diffusion coefficients for the various elements, in (µm)2 s−1 , are pro-
vided by the following formulae, where the absolute temperature T is in K
(Oikawa(26) ):


 DO = 2.44 105 exp(−11078/T ),

DMn = 7.56 107 exp(−26983/T ),
(B.1)

 D = 7.35 107 exp(−26439/T ),
 Si
DAl = 5.15 10 exp(−29566/T ).
8

Solubility products for the various precipitates, in (ppm)n (where n is

the total number of atoms in each molecule of precipitate), are given by

 log10 KMnO = 10.95 − 10830/T ,


 log10 KSiO2 = 16.00 − 22010/T ,
log10 KMn2 SiO4 = 38.56 − 46870/T , (B.2)

 = −
 10 Al2 O3
 log K 28.55 49600/T ,
log10 KMnAl2 O4 = 39.88 − 62944/T .
These formulae are taken from the Thermodata(23) bank, except for that
for KMn2 SiO4 which was adjusted using experimental data provided by
Eriksson et al.(27) The atomic masses of elements and molecular masses
of precipitates, in g(mol)−1 , are finally recalled for completeness:
MO = 15.999; MMn = 54.938; MSi = 28.086; MAl = 26.982; (B.3)
2 Equation (A.2) is based on the implicit assumption that the partial pressures of water
vapor and hydrogen are almost the same at the temperatures T and DP , in spite of the
fact that chemical equilibrium between water vapor and gaseous oxygen and hydrogen is
different at these two temperatures. This approximation is justified by the fact that the par-
tial pressure of oxygen is much lower than those of both water vapor and hydrogen.
Numerical Simulation of Internal Oxidation of Steels 155

MMnO = 70.937; MSiO2 = 60.084; MMn2 SiO4 = 201.958;

MAl2 O3 = 101.961; MMnAl2 O4 = 172.898. (B.4)

APPENDIX C: THE WAGNER SOLUTION FOR A HIGHLY

OXIDIZABLE ELEMENT
Wagner(1) considers isothermal internal oxidation of a single element
(B) by oxygen (O). A single precipitate with chemical formula Bm Op is
formed. The solubility product of this precipitate is supposed to be so
extremely small that the two elements cannot coexist in their dissolved
form. Precipitation then occurs only on some “precipitation front”, the
position of which is denoted ξ(t). In the region 0 ≤ x < ξ(t), the molar
dissolved concentration of oxygen, O ≡ CO /MO , is non-zero but the molar
dissolved concentration of the oxidizable element, B ≡ CB /MB is zero. The
converse is true in the region x > ξ(t).
The governing equations are as follows. The diffusion equations read

∂O ∂ 2O
(x, t) = DO 2 (x, t) for 0 < x < ξ(t);
∂t ∂x
∂B ∂ 2B
(x, t) = DB 2 (x, t) for x > ξ(t). (C.1)
∂t ∂x

The conditions at the precipitation front read

∂O ∂B
O(ξ(t), t) = B(ξ(t), t) = 0, −mDO (ξ(t)− , t) = pDB (ξ(t)+ , t).
∂x ∂x
(C.2)

The second condition here expresses the “stoechiometry” of the fluxes of

O and B at the precipitation front, arising from the fact that these fluxes
are entirely “consumed” by precipitation there. Initial conditions read

O(x, 0) = 0 for x > 0; B(x, 0) = B core for x ≥ 0, (C.3)

where B core is the molar core concentration of B (nominal concentration

defined by the grade of the steel). Boundary conditions read
156 Huin, Flauder, and Leblond

O(0, t) = O surf , B(+∞, t) = B core for t > 0, (C.4)

where O surf is the molar concentration of oxygen prescribed on the sur-

face of the sheet.
To solve these equations, define the following dimensionless parameters:

φ ≡ DO /DB ; µ ≡ p/m. (C.5)

Also, write the position ξ(t) of the precipitation front in the form

ξ(t) = 2γ DO t, (C.6)

where γ is an ancillary dimensionless unknown. Solving the diffusion

Equation (C.1) with the initial conditions (C.3), the boundary conditions
(C.4) and the conditions (C.2)1 at the precipitation front, one gets the fol-
lowing expressions of the concentrations:
 √
 O(x,t) = 1 − erf (x/(2 DO t)) for 0 ≤ x ≤ ξ(t),
O surf erfr
√ (C.7)
 B(x,t)
core = 1 −
erfc(x/(2 DB t))
√ for x ≥ ξ(t).
B erfc(γ φ)

Condition (C.2)2 at the precipitation front then yields the following equa-
tion on γ , which implicitly provides the value of this parameter:

exp(γ 2 )erf γ O surf

√ √ = . (C.8)
φ exp(γ 2 φ)erfc(γ φ) µB core

Finally, the molar fraction of element B in precipitate form, B prec , can be

obtained from the fluxes of O and B at the precipitation front; it is found
to be uniform within the oxidized zone 0 < x < ξ(t) and given by

B prec exp(−γ 2 φ)
core
= √ √ . (C.9)
B γ πφ erfc(γ φ)

APPENDIX D: EXTENSION OF WAGNER’S SOLUTION TO SOME

MODERATELY-OXIDIZABLE ELEMENT
The Wagner solution was extended by Huin et al.(8) to the case where
the solubility product K ≡ KBm Op of the precipitate is not very small.
The presentation in their original paper was compact and is repeated with
more details here. The new solution is only semi-analytic. The analytic
treatment reduces the problem to complex, non-linear 2nd-order ordinary
differential equations, which must be solved numerically.
Numerical Simulation of Internal Oxidation of Steels 157

When the solubility product of the precipitate is non-zero, the two

elements can coexist in their dissolved form. Precipitation is no longer
confined to some precipitation front and occurs over a whole domain.

The Extended Solution

The treatment is based on the assumption that the equation

[B(x, t)]m [O(x, t)]p = K, (D.1)

where O and B denote the molar dissolved concentrations of oxygen and

the oxidizable element like in Appendix C, is satisfied for all x and t. This
means replacing the true initial condition of zero dissolved concentration
of oxygen by the condition

O(x, 0) = O core ≡ [K(B core )−m ]1/p for x > 0. (D.2)

Full numerical simulations performed with the true initial condition and
the modified one (D.2) have yielded very similar results. This shows that
this approximation is acceptable. The physical explanation is that the ficti-
tious non-zero value O core of the deep concentration of oxygen is always
much smaller in practice than that, O surf , imposed on the surface of
the sheet. Condition (D.1) implies that precipitation occurs everywhere,
although it is negligibly small deep in the sheet.
The diffusion equations (4) read, using molar fractions O ≡ CO /MO ,
B ≡ CB /MB , P ≡ PBm Op /MBm Op instead of mass fractions:

 ∂O (x, t) = DO ∂ 2 O (x, t) − p ∂P (x, t),
∂t ∂x 2 ∂t
(D.3)
 ∂B (x, t) = DB ∂ 2 B (x, t) − m ∂P (x, t).
∂t ∂x 2 ∂t

Elimination of ∂P /∂t between these equations yields

   
∂O ∂ 2O ∂B ∂ 2B
m (x, t) − DO 2 (x, t) = p (x, t) − DB 2 (x, t) . (D.4)
∂t ∂x ∂t ∂x

The initial condition for oxygen is the modified one, (D.2), and that for
element B is the same as before, (C.3)2 . The boundary conditions (C.4)
still hold, but are now supplemented by the following ones, which result
from Eq. (D.1):

O(+∞, t) = O core , B(0, t) = B surf ≡ [K(O surf )−p ]1/m for t > 0. (D.5)
158 Huin, Flauder, and Leblond

To solve these equations, define the following “reduced” variables and

functions:
x x O(x, t) B(x, t)
u≡ √ ; v≡ √ ; X(u) ≡ ; Y (v) ≡ . (D.6)
2 DO t 2 DB t O surf B surf
By Eq. (D.1),

[X(u)]p [Y (v)]m = 1. (D.7)

Using this identity to express either Y (v) in terms of X(u) or X(u) in

terms of Y (v) in Eq. (D.4), one gets the two following equivalent differ-
ential equations:
   
µ2 B surf −1−µ  µ2 B surf −1−µ
1+ X X +2 1+ X uX 
φO surf O surf
µ2 B surf
− (1 + µ)X −2−µ X 2 = 0; (D.8)
φO surf
   
φO surf −1− µ1  O surf −1− µ1
1 + 2 surf Y Y + 2 1 + 2 surf Y vY 
µ B µ B

φO surf 1 1
− 2 surf 1 + Y −2− µ Y 2 = 0 (D.9)
µ B µ

where the notations φ, µ introduced in Eqs. (C.5) are again used. These
equations are too complex to be solved analytically, but this can be done
numerically, using for instance the classical Runge–Kutta method of order
4. When doing so, one must also fulfill the following conditions, which fol-
low from the definitions (D.6)3 , (D.6)4 of the functions X(u), Y (v) and the
boundary conditions (C.4), (D.5):

X(0) = 1; X(+∞) = O core /O surf ,
(D.10)
Y (0) = 1; Y (+∞) = B core /B surf .

If one integrates, for instance, the differential equation on the function

X(u) from u = 0 to u = +∞, one must impose X(0) = 1 and numerically
adjust the value of X (0) so as to get the correct value of X(+∞).
To calculate the molar fraction of element B in precipitate form,
B prec , note that by Eqs. (D.3) and (D.6),
  
∂B prec ∂P  m D ∂ 2 O (x, t) − ∂O (x, t) = O surf X +2uX ,
O 2 4t
(x, t) = m (x, t) = p 2 ∂x ∂t µ
∂t ∂t  DB ∂ B (x, t) − ∂B (x, t)  
= B surf Y +2vY ,
∂x 2 ∂t 4t
Numerical Simulation of Internal Oxidation of Steels 159

which yields upon integration two equivalent expression of B prec :

  surf  surf  +∞
  
 t O X +2uX X (u) 
 0 µ 4t dt = O2µ √x u + 2X (u) du,
2 DO t   (D.11)
B prec (x, t) =  t Y  +2vY 

B surf +∞ Y  (v)

 0B surf dt = + 2Y  (v) dv.
4t 2 √x
v
2 DB t

Reduction to the Wagner Solution in the Limit K→ 0

The reduction of the solution just presented to the Wagner solution
in the limit K → 0 is not trivial and must therefore be explained in detail.
If K = 0, Eq. (D.1) reduces to O = 0 or B = 0. In practice, O is zero
on the interval [ξ(t), +∞) and B is zero on the interval [0, ξ(t)].
Consider now concentration O. By Eq. (D.5)2 , K = 0 implies B surf =
0. Discarding therefore terms proportional to B surf /O surf in Eq (D.8), one
gets

X  + 2uX  = 0. (D.12)

In fact, because of the presence of terms proportional to X −1−µ and

X−2−µ in Eq. (D.8), this simplified equation holds only if X is non-zero,
that is for 0 < x < ξ(t), or equivalently 0 < u < γ (by the definitions (C.6)
and (D.6)1 of γ and u). Solving Eq. (D.12) and accounting for the con-
ditions X(0) = 1 and X(γ ) = 0 (equivalent to the condition O(ξ(t), t) = 0
defining ξ(t)), one gets
erf u
X(u) = 1 − for 0 ≤ u ≤ γ , (D.13)
erf γ
and this is equivalent to Eq. (C.7)1 for O(x, t).
Consider now concentration B. The use of the “reduced” function Y
in the differential equation (D.9) raises a problem in the limit K → 0 since
the very definition (D.6)4 of this function implies that it becomes infinite
then. To solve this problem, take K > 0, substitute B/B surf for Y in this
equation and multiply it by B surf ; the result is
 
φ surf surf µ1 −1− µ1  1 surf surf µ1 −1− µ1
1 + 2 O (B ) B B + 2 1 + 2 O (B ) B υB 
µ µ

φ 1 1 1
− 2 O surf (B surf ) µ 1 + B −2− µ B 2 = 0, (D.14)
µ µ
where B is considered as a function of v. It is then easy to take the limit
1
K → 0 by discarding terms proportional to (B surf ) µ ; one thus gets

B  + 2vB  = 0. (D.15)
160 Huin, Flauder, and Leblond

1
Again, because of the presence of terms proportional to B −1− µ and
1
B −2− µ in Eq. (D.14), this simplified equation
√ holds only if B is non-zero,
that is for x > ξ(t), or equivalently v > γ φ (by the definitions (C.5)1 ,
(C.6) and (D.6)2 of φ, γ and v). The√solution of equation (D.15), with the
conditions B(+∞) = B core and B(γ φ) = 0 (equivalent to the condition
B(ξ(t), t) = 0), reads
B(v) erfcv 
core
=1− √ for v ≥ γ φ, (D.16)
B erfc(γ φ)
and this is equivalent to Eq. (C.7)2 for B(x, t).
The next task is to establish, in the limit K → 0, Eq. (C.8) which pro-
vides the value of γ . Since it is a consequence of the expressions of O and
B and condition (C.2)2 , it suffices to show that this condition holds. This
can be done by integrating Eq. (D.4) between x −  and x + , then letting
K → 0 while taking x ≡ ξ(t), and finally letting  → 0, accounting for the
finiteness of ∂O/∂t and ∂B/∂t.
The last task is to check that any of the equivalent expressions (D.11)
of B prec reduces to (C.9) in the limit K → 0. The differential equation
(D.8) on X then reduces to (D.12), so that it seems at first sight that the
integrand in equation (D.11)1 is zero. However in the limit K → 0, the
function X has a discontinuous slope at the point u = γ , since X  (γ + ) is
zero but X (γ − ) is not. As a consequence, the limit of X  has a com-
ponent of the form −X (γ − )δ(u − γ ) where δ is Dirac’s function, so that
the limit of X  (u)/u + 2X  (u) is not simply zero but −X  (γ − )δ(u − γ )/u.
Therefore equation (D.11)1 becomes
 +∞
B prec (x, t) 1 X  (γ − )δ(u − γ )
= − du
O surf 2µ √x u
2 DO t
 √
1 X (γ − )
− 2µ if x < 2γ DO t,
= γ √ (D.17)
0 if x > 2γ DO t.
Upon use of Eqs. (D.13) and (C.8), expression (D.17)1 becomes
O surf 2 exp(−γ 2 ) exp(−γ 2 φ)
B prec = √ = B core √ √ ,
2µ π γ erf γ γ π φ erfc(γ φ)
and this is identical to equation (C.9).

Numerical Example and Comments

We shall illustrate the predictions of the extended model by show-
ing results corresponding to internal oxidation of a 1.235% Mn (in
Numerical Simulation of Internal Oxidation of Steels 161

weight) steel. The precipitate formed is MnO. The DP of the anneal-

ing atmosphere is −40 ◦ C and the temperature is 800 ◦ C. The values of
the diffusion coefficients and the superficial concentration of oxygen are
then DO = 8.01 (µm)2 s−1 , DMn = 9.06 10−4 (µm)2 s−1 and CO surf = 9.23 ×
−3
10 ppm. Several values of the solubility product KMnO are envisaged: the
experimental value is K = 7.19 (ppm)2 , but the values 3 K, 0.001 K and
0 (Wagner’s case) are also considered in order to illustrate the influence of
this parameter. Results are given at time t = 60 s.
Figures D.1 and D.2 show the computed mass concentrations of dis-
solved O and Mn, together with those of the classical Wagner solution.
One sees that for KMnO = 0.001K, the numerical profile of dissolved Mn
completely agrees with that in the Wagner solution, and the same is true
of the profile of dissolved O except for small differences very close to
the precipitation front. (A better agreement could be obtained by using
a smaller value of KMnO in the numerical simulation). For higher values
of KMno , however, there are large differences; observe for instance that in
the numerical simulations, the concentrations of dissolved O and Mn are
simultaneously non-zero everywhere, unlike in the Wagner solution.
Figure D.3 shows the mass fraction of oxidized Mn in a similar way.
Again, for KMnO = 0.001K, the numerical solution is rather close to that
of Wagner except near the precipitation front, but here the difference is
quite conspicuous even for such a small value of KMnO . For larger val-
ues of KMnO , the fraction of oxidized Mn increases towards the surface.

Fig. D.1. Semi-analytical solution—fraction of dissolved O.

162 Huin, Flauder, and Leblond

Fig. D.2. Semi-analytical solution—fraction of dissolved Mn.

Fig. D.3. Semi-analytical solution—fraction of oxidized Mn.

This behavior is in clear contrast with the Wagner solution and must be
explained.
Since
√ in the case considered, the solution depends only on the var-
iable x/ t, understanding the dependence of the fraction of precipitate
with respect to x is equivalent to understanding its dependence with
respect to t. In the Wagner solution, at a given point, the fraction of
Numerical Simulation of Internal Oxidation of Steels 163

precipitate is constant in time once the precipitation front has swept the
point considered, because precipitation no longer occurs. By the equiva-
lence of x and t, this implies that this fraction must also be independent
of position within the oxidized zone at a given instant. In contrast, in
the extended solution discussed here, precipitation is no longer confined
to some localized front but occurs over a whole domain. Thus, at fixed
x, precipitation is not instantaneous but lasts for some time, so that the
fraction of precipitate must continuously increase with t. Again, by the
equivalence of x and t, this implies that for fixed t, it must increase for
decreasing x.

APPENDIX E: EXTENSION OF THE SOLUTION TO SEVERAL

OXIDIZABLE ELEMENTS
Huin et al.’s(8) solution presented in Appendix D was for a single
oxidizable element. However, they also developed it for two oxidizable
elements. It could in fact be extended to an arbitrary number of such
elements, including even the possible formation of “mixed” oxides (i.e.,
involving several oxidizable elements each), the sole, albeit admittedly con-
siderable restriction being that the sequence, in time or space, of successive
combinations of oxides must be known a priori.
We consider here only the case of two oxidizable elements, B and
C. Two oxides, Bm Op and Cn Oq , with solubility products KBm Op ≡
KB , KCn Oq ≡ KC , can be formed. Oxide Bm Op is assumed to be more sta-
ble than oxide Cn Oq for the core values B core , C core of the concentrations
of B and C defined by the grade of the steel. It is also assumed, like in
Appendix D, that the more stable oxide Bm Op is always and everywhere in
equilibrium with elements O and B in solution in the matrix. This means
that the equation
[B(x, t)]m [O(x, t)]p = KB , (E.1)
(where O and B again denote the molar concentrations of oxygen and the
oxidizable element B) holds for all x and t. Since Cn Oq is less stable than
Bm Op , the core concentration of oxygen which results from Eq. (E.1),
 1/p
O core ≡ KB (B core )−m , (E.2)
is insufficient to oxidize element C. Thus oxide Cn Oq is present only in
some interval [0,ζ (t)) and absent in the interval [ζ (t), +∞), which means
that
[C(x, t)]n [O(x, t)]q = KC for 0 ≤ x ≤ ζ (t);
[C(x, t)]n [O(x, t)]q < KC for x > ζ (t), (E.3)
164 Huin, Flauder, and Leblond

where C denotes the molar concentration of element C.

The diffusion equations (4) read, using molar fractions O ≡ CO /MO , B ≡
CB /MB , C ≡ CC /MC , P ≡ PBm Op /MBm Op , Q ≡ PCn Oq /MCn Oq instead of mass
fractions:

 ∂2O ∂Q
 ∂t (x, t) = DO ∂x 2 (x, t) − p ∂t (x, t) − q ∂t (x, t),
∂O ∂P

2
∂t (x, t) = DB ∂x 2 (x, t) − m ∂t (x, t),
∂B ∂ B ∂P (E.4)


 ∂C (x, t) = D ∂ C (x, t)
2 ∂Q
− n (x, t).
∂t C ∂x 2 ∂t

Elimination of ∂P /∂t and ∂Q/∂t between these equations yields

   
∂O ∂ 2O ∂B ∂ 2B
mn (x, t)−DO 2 (x, t) = np (x, t)−DB 2 (x, t)
∂t ∂x ∂t ∂x
 
∂C ∂ 2C
+mq (x, t)−DC 2 (x, t) .
∂t ∂x
(E.5)

The initial conditions are

O(x, 0) = O core for x > 0; B(x, 0) = B core , C(x, 0) = C core for x ≥ 0. (E.6)

Boundary conditions read


 O(0, t) = O surf , O(+∞, t) = O core ,
B(0, t) = B surf surf −p
≡ [KB (O ) ] , B(+∞, t) = B core for
1/m t > 0.

C(0, t) = C surf ≡ [KC (O surf )−q ]1/n , C(+∞, t) = C core
(E.7)

All concentrations and fluxes must in addition be continuous at the point

ζ (t).
To solve these equations, define the dimensionless parameters

φ ≡ DO /DB ; ψ ≡ DO /DC ; µ ≡ p/m; ν ≡ q/n (E.8)

and the “reduced” variables and functions

 

 u ≡ 2√xD t 
X(u) ≡ O(x, t)/O ,
surf
 O
v≡ 2 D t;
√x
Y (v) ≡ B(x, t)/B ,
surf (E.9)

 B 
Z(w) ≡ C(x, t)/C surf .
w ≡ √x
2 DC t

Also, write ζ (t) in the form


ζ (t) = 2χ DO t, (E.10)
Numerical Simulation of Internal Oxidation of Steels 165

where χ is an ancillary dimensionless unknown. Equations (E.1) and

(E.3)1 then read

[X(u)]p [Y (v)]m = 1 for all u ; [X(u)]q [Z(w)]n = 1 for u ≤ χ.

(E.11)

Using these equations to express Y (v) and Z(w) in terms of X(u) in Eq.
(E.5), one gets the following differential equation on the function X in the
region 0 < x < ζ (t):
 
µ2 B surf −1−µ ν 2 C surf −1−ν
1+ X + X X
φO surf ψO surf
 
µ2 B surf −1−µ ν 2 C surf −1−ν
+2 1 + X + X uX
O surf O surf
 
µ2 B surf ν 2 C surf
− (1+µ)X−2−µ + (1 + ν)X −2−ν X 2 = 0 for 0 < u < χ .
φO surf ψO surf
(E.12)

In the region x > ζ (t), ∂Q/∂t is zero so that Eqs. (E.4)3 and (E.5) similarly
yield
   
µ2 B surf −1−µ µ 2 B surf
1+ X X  + 2 1 + X −1−µ uX 
φO surf O surf
µ2 B surf
− (1 + µ)X −2−µ X 2 = 0 for u > χ; (E.13)
φO surf

Z  + 2wZ  = 0 for w > χ ψ. (E.14)

The solution of Eq. (E.14) is

√ 
 π  
Z(w) = Z(χ ψ) + exp(χ 2 ψ) erfw − erf(χ ψ)
 2 
×Z  (χ ψ) for w ≥ χ ψ. (E.15)

For fixed values of X  (0) and χ , the concentrations may be obtained as fol-
lows: get the function X(u) by integrating numerically Eq. (E.12) from 0 to
χ using the condition X(0) = 1 and the specified value of X  (0), then Eq.
(E.13) from χ to +∞ using the continuity of X and X at the point u = χ ;
get next the function Y (v) from Eq.√(E.11)1 ; finally get the function Z(w)
√
from equation (E.11)2 for 0 ≤ w ≤ χ ψ, and from Eq. (E.15) for w ≥ χ ψ.

The problem is thus reduced to finding the correct values of X (0) and χ .
166 Huin, Flauder, and Leblond

These values must be adjusted numerically so as to find the correct values

of X(+∞) and Z(+∞), namely O core /O surf and C core /C surf .
The molar fractions P , Q of precipitates Bm Op and Cn Oq can finally
be obtained in the same way as in the case of a single precipitate: by Eqs.
(E.4)2,3 ,
  
 ∂P (x, t) = 1 DB ∂ 2 B2 (x, t) − ∂B (x, t) = B surf Y  +2vY  ,
∂t m ∂x ∂t  m 4t
 ∂Q (x, t) = 1 DC ∂ 2 C2 (x, t) − ∂C (x, t) = C surf Z  +2wZ  ,
∂t n ∂x ∂t n 4t

so that
 t Y  +2vY 
 +∞  Y  (v) 
 (v) dv,


P (x,t)
surf = 0 = 2m
4mt dt
1
√ +2Y

 B
x
 2 D√B t 
v

 t Z  +2wZ   1 χ x ψ Z  (w) +2Z  (w) dw if x < ζ (t),

Q(x,t)
= 0 4nt dt = 2n 2√DC t w

 C surf 0 if x ≥ ζ (t),
(E.16)
where Eq. (E.14) has been used.

REFERENCES
1. C. Wagner, Zeitschrift fur Elektrochemie 63, 772 (1959) (in German).
2. R. Rapp, Corrosion 21, 382 (1965).
3. J. S. Kirkaldy, Canadian Metallurgical Quarterly 8, 35 (1969).
4. D. P. Whittle, F. Gesmundo, B. D. Bastow, and G. C. Wood, Oxidation of Metals 16, 159
(1981).
5. H. J. Christ, W. Christl, and H. G. Sockel, Werkstoffe und Korrosion 37, 385 (1986) (in
German).
6. F. Gesmundo and B. Gleeson, Oxidation of Metals 44, 211 (1995).
7. F. Gesmundo, F. Viani, and Y. Niu, Oxidation of Metals 45, 51 (1996).
8. D. Huin, V. Lanteri, D. Loison, P. Autesserre, and H. Gaye, in Microscopy of Oxidation
- 3, Newcomb and Little, eds. (the Institute of Metals, London, (1997) pp. 573–586.
9. F. Gesmundo, F. Viani, and Y. Niu, Oxidation of Metals 47, 355 (1997).
10. F. Gesmundo, P. Castello, F. Viani, and C. Roos, Oxidation of Metals 49, 237 (1998).
11. F. Gesmundo and Y. Niu, Oxidation of Metals 51, 129 (1999).
12. Y. Niu and F. Gesmundo, Oxidation of Metals 56, 517 (2001).
13. D. L. Douglass, Oxidation of Metals 44, 81 (1995).
14. K. Bongartz, D. F. Lupton, and H. Schuster, Metallurgical Transactions A 11, 1883
(1980).
15. E. Rank and U. Weinert, IEEE Transactions on Computer-Aided Design of Integrated
Circuits and Systems 9, 543 (1990).
16. R. Fortunier, J. B. Leblond, and D. Pont, Phase Transformations During the Ther-
mal/Mechanical Processing of Steel, Hawbolt and Yue, eds., (the Canadian Institute of
Mining, Metallurgy and Petroleum, Montreal, 1995), pp. 357–371.
17. J. Crank, The Mathematics of Diffusion (Clarendon Press, Oxford, 1967).
18. Ph. Autesserre, Modélisation du phénoméne d’oxydation interne : présentation d’un
modéle aux différences finies. IRSID Internal Report R100.05 (in French) (2000).
19. D. Huin and J. Lehmann, Modélisation de l’oxydation sélective lors du recuit: point
d’avancement du code Différences Finies. IRSID Internal Report PCMO/01/N/092/PC
(in French) (2001).
Numerical Simulation of Internal Oxidation of Steels 167

20. P. Flauder and D. Huin, Oxydation sélective des aciers lors du recuit. IRSID Internal
Report PC&S/02/N/114/GC (in French) (2002).
21. J. B. Leblond, Nonlinear Analysis: Real World Applications 6, 297 (2005).
22. D. Loison, D. Huin, V. Lanteri, J. P. Servais, and R. Cremer, Proceedings of the
5th International Conference on Zinc and Zinc Alloy Coated Steel Sheet (GALVA-
TECH’2001), Brussels, (2001), pp. 203–209.
23. Thermodata. Electronic data bank for thermodynamic quantities, available online at the
address http://thermodata.online.fr (2005).
24. A. Rist, M. F. Ancey-Moret, C. Gatellier, and P. V. Riboud, Equilibres thermodynami-
ques dans l’élaboration de la fonte et de l’acier. Techniques de l’Ingénieur, traité Matéri-
aux Métalliques, M 1730 (in French) (1974).
25. J. H. Swisher and E. T. Turkdogan, Transactions of the Metallurgical Society of AIME
239, 426 (1967).
26. H. Oikawa, Transactions of the Iron and Steel Institute of Japan 68, 1489 (1982).
27. G. Eriksson, P. Wu, M. Blander, and A. D. Pelton, Canadian Metallurgical Quarterly 33,
13 (1994).