Acta mater. Vol. 44, No. 11, pp.

4375-4385,1996
Copyright 0 1996Acta Metallurgica Inc.
Pergamon Published by Elsevier Science Ltd
PII 81359-6454(96)00084-5 Printed in Great Britain. All rigbts reserved
1359-6454/96$15.00+ 0.00

A NUMERICAL SIMULATION OF THE JOMINY

END-QUENCH TEST

D. HijMBERG
Weierstrass Institute for Applied Analysis and Stochastics, MohrenstraBe 39, D-101 17 Berlin, Germany

(Received 3 April 1995; accepted 22 February 1996)

Abstract-We present a numerical algorithm for simulating the Jominy end-quench test and deriving
continuous cooling diagrams. The underlying mathematical model for the austenite-pearlite phase
transition is based on Scheil’s Additivity Rule and the Johnson-Mehl equation. For the formation of
martensite we compare the Koistinen-Marburger formula with a rate law, which takes into account the
irreversibility of this process. We carry out numerical simulations for the plain carbon steels C 1080 and
C 100 W 1. The results suggest that the austenite-pearlite phase change may be described decently by
the Additivity Rule, except for the incubation time. On the other hand, using a rate law to describe the
martensite formation is preferable to the Koistinen-Marburger formula, which leads to unphysical
oscillations of the cooling curves in simulated CCT-diagrams. Copyright 0 1996 Acta Metallurgica Inc.

1. INTRODUCTION this phase transition, where the Koistinen-Marburger

formula was replaced by a rate law, accounting for the
In this paper we describe a mathematical model for
irreversibility of the martensite formation.
the phase transitions in eutectoid carbon steel and use
Here we present a numerical realization of this
it to develop a numerical scheme for the simulation
model and use it to simulate hardenability curves for
of the Jominy end-quench test.
two different plain carbon steels. In Section 2 we
In this test a cylindrical steel bar is heated up to its
briefly review the mathematical model as described in
austenitic state. Then it is put in a fixation and
[16]. In Section 3 we discuss the numerical
quenched by spraying water on its lower end (cf.
implementation of the model. Finally, in Section 4 we
Fig. 1). Afterwards the hardness is measured at
discuss the results of the numerical calculations.
increasing distances from the quenched end. The
results are plotted in a hardenability curve. It serves
as a measure for the hardness penetration depth of 2. THE MATHEMATICAL MODEL
this steel and thereby defines its range of application.
2.1. Time-temperature transformation diagrams
For a simulation of the Jominy test one first needs
a mathematical model to describe the growth of In eutectoid carbon steel two phase transitions may
pearlite and martensite as well as recalescence effects occur: one from austenite to pearlite and one from
in the steel bar owing to the latent heat of the phase austenite to martensite. The A-P transformation is
changes. driven by the diffusion of carbon atoms, it is time-
A lot of work has been spent on simulating phase dependent and irreversible. The A-M transformation
transitions in steel, e.g. [1], [7], [13], [14] and [19]. The is diffusionless. It is temperature-dependent in such a
first mathematical investigation of phase transitions in way that the fraction of martensite only increases
steel has been carried out by Visintin [26], but he only during non-isothermal stages of the cooling process.
considered the austenite-pearlite transformation. The evolution of the phase transitions is usually de-
Based on this model Verdi and Visintin [25] suggested scribed in time-temperature transformation diagrams.
a numerical scheme for simulating the austenite- Figure 2 depicts an isothermal-transformation (IT)
pearlite phase change, without presenting numerical diagram for the plain carbon steel C 1080. Here, A,
results. In [15] the author developed a model for the and M, denote the starting temperatures for the
austenite-pearlite and the austenite-martensite phase formation of pearlite and of martensite, respectively.
change that is based on Scheil’s Additivity Rule and For fixed temperatures the bold-faced curved lines
the Koistinen-Marburger formula. It turned out that indicate the beginning of the austenite-pearlite
the Koistinen and Marburger formula is an transformation, i.e. the time when 1% of the
insufficient tool for simulating the growth of austenite has been transformed, and the end of the
martensite, since it does not take care of the transformation, i.e. the time when 99% of the
irreversibility of this transition. This led to unreason- austenite has been transformed.
able oscillations in the simulated CCT diagrams. Then In the non-isothermal case the phase evolutions are
in [ 161 the present author investigated a new model for represented in a continuous-cooling-transformation
4376 HijMBERG: A NUMERICAL SIMULATION OF THE JOMINY END-QUENCH TEST

(CCT) diagram. This can be derived from an

isothermal-transformation diagram by superimpos-
ing several cooling curves on it. On each curve the
beginning and the end of the transformation are
marked. Then the connection of the respective points
defines the CCT diagram. Compared to an IT
diagram the transformation curves are moved to later
time and lower temperature (cf. Fig. 3).

2.2. The austenite-pearlite phase change

As the A-P transformation is a nucleation and
growth process, it is governed by the nucleation rate
(the amount of nuclei of the new phase formed per
unit time and volume) and by the growth rate of the
nuclei. Assuming these rates to be constant and
furthermore spherical growth of the nuclei, Johnson 0.5 I 2 5 D IO’ d 106

and Mehl[21] in 1939 derived the equation: TIME - SECONDS

P(t) = 1 _ e-77/3fiG’r Fig. 2. Isothermal-transformation diagram for the plain

(1)
carbon steel C 1080 (from [2]).
for the A-P transformation in the isothermal case.
Here p is the fraction of pearlite, ?? is the nucleation
rate and G is the growth rate of the nuclei. Here z(T,p) denotes the time to transform the
As in [l] and [26] we use (1) in the parametric fraction p to pearlite at constant temperature T.
version: Thus, by equation (2):

p(t) = 1 - e-*(Q@‘. (2)

z(T,p) = (--WY”. (4)
The temperature dependent coefficients a(T) and
b(r) can easily be calculated using the transformation
Equation (3) was derived by Scheil[24] to predict the
curves in the IT diagram (cf. Section 2.3).
incubation period of the A-P transformation. Later
In the non-isothermal case, we use the additivity
Avrami [5] and Cahn [S] showed that equation (3) can
rule to describe the formation of pearlite:
be applied to characterize the kinetics of a class of
phase changes which they called additive.
1
I0 V(5), p(t)) dr = l.
(3)
Although the pearlite phase change is not an
additive transformation in their sense (cf. [9]),
according to a comparative investigation by Hayes
[12] the additivity rule is a better tool for predicting

Deviceforltdng and
piece
csnlrhg maIesl

t4co

t-- Test piace in posilbn

1200

tom
E
em
i
b6w

400

200

Fig. 3. Derivation of a continuous-cooling from an

Fig. 1. Diagram of the cooling device (from 1201). isothermal-transformation diagram (from [4]).
 HOMBERG: A NUMERICAL SIMULATION OF THE JOMINY END-QUENCH TEST 4377

3.5

0.02
3

2.5

1.5

0.005

loo 200 300 400 500 600 700 8cQ 0 109 200 300 400 500 600 700 800
T T

Fig. 4. The data functions a(T) and b(n for the carbon steel C 1080.

the course of the phase change than a rate law. Simple manipulations show that the solution is
Moreover, measurements by Hawbolt et al. [13] show given by:
that also in quantity the A-P transformation is
a(T) = ln(ln(O.01)) - In(ln(0.99)
described well by the additivity rule, except for the (6a)
incubation period where the pearlite fraction ln(tXr)) - ln(WN
predicted by the additivity rule shows only poor b(T) = -ln(O.Ol)t,(T)-“‘n. (6b)
coincidence with the measurements. It should be
noticed that equations of this type are also used for Figure 4 shows the behaviour of the coefficients
a(T) and b(T) in the temperature range (Mi, A.,) for
modeling fatigue effects, e.g. the Palmgren-Minor
the plain carbon steel C 1080. For temperatures
rule (cf. [6]).
A different approach to model a nucleation and below M,? the values are obtained by linear
extrapolation. Although it cannot be concluded from
growth process was chosen by Andreucci et al. [3].
the diagram, b also stays positive in the whole
Going back to the ideas of Johnson and Mehl they
derived an integral equation to describe the temperature range. Hence, in the following we
solidification of polymers in the non-isothermal case. assume a and b to be continuous functions on
[M/-, A,], bounded away from zero.

2.3. Identifying coefjicients from IT diagrams 2.4. An initial value problem for the A-P transform-
ation
Assuming that the generalized Johnson-Mehl
equation (2) appropriately describes the isothermal A simple way to exploit the additivity rule is to
evolution of the phase fractions we present a simple differentiate it formally with respect to time. Since we
method to obtain the data functions a(T) and b(T) obtain an inner derivative p, rearranging terms leads
from the IT diagrams. to the following initial value problem:
Since the bold-faced curves in these diagrams are
P(O) = PO (74
the “is0-fractions” p = 0.01 and p = 0.99, we
interpret these transformation curves as the ii(t) =fCt> P(t), TL (7b)
respective graphs of functions:
with

t,: [MT, A,l+IW+, tr: [MY,A,l+IW+,

which measure the beginning and end of the pearlite

.?@>P(f)> n =-
(s’a
cl p
dr
z(T(t)>p(t))
-’r(T(t;
,
p(t)).

transformation for given temperature. These data (8)

functions can be drawn from the IT diagram. Then
It may be proved (cf. [15]) that for any given
the wanted coefficients are the solution to the
(integrable) temperature evolution T: [0, tE]-+[M,, A,]
following nonlinear system of equations:
equations (7a and b) admit a unique solution p,
satisfying:
0.01 = 1 - e-mf” W
0 <p(t) < c,, < 1, for all te[O, tE], (9)
0.09 = 1 - emb(nrT”. (5a) with a constant c,,, depending only on the end time
4378 HijMBERG: A NUMERICAL SIMULATION OF THE JOMINY END-QUENCH TEST

tE. Moreover, we have:
P(l) > 0,
i.e. the irreversibility of the austenite-pearlite
transformation carries over to the model. Unfortu-
nately, as Fig. 4 shows, the coefficient a, which was
equal to 4 in the original Johnson-Mehl equation and
assumed to be greater than 1 in [25] and [26], actually
takes values less than 1 if the temperature is in a range
just below A,. In this case, we can prove the following:
Proposition 1. Let T: [0, tE]+[M,, A,] be a continuous
function, such that
a(T(t)) < 1 for all 0 < t < tl
then the following are valid:
lim p(t) = 0,
r-0(+1
lim e(t) = co,
1-o(+)
For the proof, we refer to [15].
In a nucleation and growth process the increase of
the volume fraction of the new phase should be
“small” during the incubation time, which is a
contradiction to equation (11 b). Thus, Proposition 1
gives the mathematical reason why the additivity rule
does not work well for the early stages of the
transformation. As said before, this fact has also been
observed experimentally.
To overcome this difficulty, we adopt the following
philosophy: We define an incubation time t,, which
we keep fixed. Giving up the aim of predicting the
exact evolution kinetics during this incubation time,
we just gauge the processs by demanding that the
additivity rule shall hold, when the end of the
incubation time is reached. This leads to the
following model:
?? Let T: [0, tE]+R be a given temperature
evolution,
?? tle(O, tE) the fixed incubation time,
depending on T,
?? p. is defined by:
prevents the formation of pearlite above the critical
temperature A,.
(10)
2.5. The austenite-martensite phase change
While the additivity rule is a well investigated
decent tool for describing the growth of pearlite,
there seems to be no satisfactory model at hand for
the martensitic transformation in steel.
Usually, exponential growth laws like the
Koistinen and Marburger formula:
m(r) = 1 - e-CM-W) (14)
are used (cf. [15], [17], [18], [22]).
These equations all have in common that they do
not model the irreversibility of the austenite-marten-
(lla) site phase transition. Thus, in numerical simulations
based on these models, owing to the release of latent
(1 lb) heat, usually a decrease in the martensite fraction is
observed (cf. [15] and Section 4).
The formation of martensite starts below the
critical temperature M, and the volume fraction of
martensite only grows during non-isothermal stages
of a cooling process. At this stage of the exposition,
where we assume the temperature evolution to be
known a priori, one could argue that growth laws
such as equation (14) are still valid, if only they are
modified by the logical statement that the volume
fraction of martensite never decreases. For instance,
one could replace equation (14) with:
m(t) = EC:; (1 - e-aMa-T(s))). (14’)
However, owing to the latent heat, the phase
transitions interact with the temperature evolution.
Therefore, it is important to keep track of the actual
transformation kinetics. Hence, we propose the
following rate law for the growth of martensite:
m(0) = 0, (W
then, ti(t) = (1 - m(t))G(T(t))H(-T!(t)). (15b)
Here, again H is the heaviside function. G shall be
bounded, positive and (Lipschitz-) continuous,

??
s
0
1
o r(T(5), PO) dr = ”

The fraction of pearlite is determined

(12)

by the
satisfying G(x) = 0 for all x 3 h4,. Putting m(0) = 0,
we tacitly assume that we start with a temperature
T(0) > M,.
If during some stage of a heat treatment cycle
following initial value problem (IVP): either T > A4, or Tis increasing, i.e. T, 3 0, according
to equation (15b) we have k(t) = 0, whence no
P(0) = Par (13a) martensite is produced during this stage.
Moreover, since ti > 0, the irreversibility of the
o<t<tr martensite transformation is now incorporated in the
p(t),T)H(A, - T(t)) :t, < t < tE. model.

(13b) 2.6. The complete model

The heaviside function In equations (13b) and (15b), actually, not the
fractions p and m occur but the volume fraction of
1, x>o austenite, which is 1 -p or 1 - m, respectively.
H(x) =
0, x,<o Therefore, to combine both models one only has to
replace these terms by the volume fraction of
 HOMBERG: A NUMERICAL SIMULATION OF THE JOMINY END-QUENCH TEST 4319

austenite in the case when both pearlite and exists a differentiable material function i? such that
martensite are present, i.e. 1 -p - m. the internal energy takes the form:
So we end up with the following initial value
problem for the phase transitions in eutectoid carbon e(x, t) = W, P, ml, (20)
steel: with the partial derivatives

P(0) = PO> (164 a6 a6 a6

-=c
3 ap = -L -=
am -L,. (21)
ar
m(0) = 0, (16b)
Here c denotes the specific heat at constant pressure
p(t) = (1 - p(t) - m(t))
and L, and L.,,, denote the latent heats of the
x f(t, p(t), m(t), I?ff(A, - T(t)), (164 austenite-pearlite and the austenite-martensite phase
change, respectively.
m(t) = (1 -p(l) - m(t)FW(t))H(-- T’,(t)), (16d) p, c, L, and L, shall not depend on the phase
fractions p, m. Thus, we obtain the following balance
where we define:
of energy:
-I

(s dt aT
.f(t, P>m, T):= -
0 4T(5N7(Wj P, m) >

ln( 1 - p - m) = B(T)&(T) z + P(T)L,(T)

ap am
at, in Q, (22)
fL(t - t,). (17)
’ W(t), p, m)
together with boundary and initial conditions:
Here, 7( T, p, m) is defined by:

-k(T) g = y(T)(T - Tr), in r x (0, tb), (23a)

ln( 1 - p - m) ““’
z(T,p, m) =
( -
b(T) > .
(18)
T(,,O)= A,, inR. (23b)
The following Proposition summarizes the properties Here, Tr is the outside temperature and y the heat
of the preceding model. exchange coefficient.
Proposition 2. Let T: [0, tr]+R be an integrable and To allow mathematical treatment of the problem,
(weakly) differentiable temperature evolution with we replace the heaviside function with the following
t?(O)= A, and t, E (0, T) the jixed incubation time. regularized version (cf. Fig. 5):
Then the following are valid
(1) p. is uniquely dejined by: 0, x < 0,
f%(x) = $X, O<x<6, (24)

s 4
1
o W(t), PO)
dr= 1. 1 I, x> 6,
where 6 > 0 is a “small” parameter. Introducing the
(2) The ZVP [equations (16aHd)] has a unique further notation Aa(. -H6( - .), and using
(absolutely) continuous solution (p, m). equations (16a)--(d) we end up with the following
(3) PO< p(t) + m(t) < c~,,~,< 1 for all t < 10, 4. nonlinear parabolic problem (P,i) for phase
See [16] for the proof and the precise formulation transitions in eutectoid carbon steel:
of the necessary assumptions.
U%(T) F+ ,G”W~(T)(l -P - mNWM(Z)
2.7. Three-dimensional case
Let R c R-’be bounded with smooth boundary -V. (k(T)VT) = ~(WdTlp,, in ~2, (254
aa = :r and Q:=CI x (0, ts).
As mechanical effects are neglected in this paper, -k(T) g = y(T)(T - T,), in r x (0, tE), (25b)
using Fourier’s law of heat conduction, we get the
following balance of energy: T(,, 0) = A,, in R. (25c)
& Here, for almost all XE~, (p(x, .), m(x, .)) is the
p-+‘(kVT)=O, (19)
solution to the following (IVP):
where p is the mass density, e the specific internal p(x, 0) = po(T(x, .)), [cf. equation (12)] (25d)
energy and k the heat conductivity of the material
under consideration. m(x, 0) = 0, We)
In a spatial model the propagation of latent heat
pl(x, t) = (1 - p(x, t) - m(x, t))
released during the phase changes has to be
considered. Following [27], it is assumed that there f(t, p(x, t), m(x, t), T(x, .))&(A, - T(x, t)), (Xf)
4380 HOMBERG: A NUMERICAL SIMULATION OF THE JOMINY END-QUENCH TEST

r&(x, t) = (1 - p(x, t) - m(x, t)) doubtful, whether one would be able to obtain
enough measurements to include this dependency in
W(x, t))ffa(- Z(x, 0). Wg) numerical simulations.

The following theorem shows that problem (Pa) is

well-posed: 3. NUMERICAL METHOD
Theorem 1. Assume that the incubation tl has been
3.1. The algorithm
chosen small enough, then (PJ) has a unique solution
(T,p, m), where the phase fractions (p, m) satisfy the In this section we will apply our model to simulate
properties of Proposition 2. the Jominy end-quench test. Owing to the symmetries
Instead of assuming the incubation time tl to be of the problem (cf. Fig. l), we make use of cylindrical
chosen “small enough’, one could also demand coordinates. Thus, we obtain the following energy
am/at = 0 a.e. in (0, tI) or pa E (0, 1) constant, balance:
independent of T.
The first case refers to a heat treatment with a
moderate cooling rate, producing pearlite and
subsequently possibly some martensite.
The second condition applies to quench cooling,
i.e. very fast cooling to achieve a nearly pure in R x (0, T), (26)
martensitic structure. In this case it is reasonable to
assume p. to be constant, because no more pearlite
will be formed during the cooling process. with R = (0, R) x (0, H), where R is the radius and
From a mathematical point of view it is interesting H the height of the steel bar. Moreover, we have used
to see what happens if the regularization parameter the abbreviations:
6 tends to zero. This question has been investigated
A(T) = ,4T)c(T) (27)
in [16], we only want to remark here that one still
obtains a solution in this case. For the proof, we had B(T) = P(T)L~TNP, m, T>
to assume that L,, L,,, are (Lipschitz) continuous,
+ dT)LdT)f2(p9 m, T), (28)
positive, bounded functions of temperature T, and
that p, c, k, y are positive constants. where f, and f2 are the right-hand sides in equations
If one is only interested in the case 6 > 0 fixed, (250 and (g).
which is clearly the most important case in view of According to Fig. 6, we consider the following
practical applications, Theorem 1 can be proved boundary conditions:
assuming that y is continuously differentiable and
p, c, k are (Lipschitz) continuous, positive, bounded rx(T- Tw), in r1 x (0, tB),
functions of temperature T. Moreover, they may in r2 x (0, tB),
depend on x and t in a rather general way.
in T3 x (0, tB),
(29)
A dependency of p and c on the phase fractions is
not covered by this theory and would require further 10, in r4 x (0, ts).
analysis. Probably it would be difficult to prove
uniqueness in this case. What is more, it seems to be Here, TW and TL denote the temperatures of water

Fig. 5. The functions Ha and AJ

 Hi)MBERG: A NUMERICAL SIMULATION OF THE JOMINY END-QUENCH TEST 4381

.z condition on r4, we obtain:

r3
H ig% 0, for r<<l.

Next, we introduce the transformation:

i=~(N,+l)+v+l, O<p<NN,, O<v<N:

(38)
and the vector T, E R’“,N = (N, + l)(N: + 111,defined
r4 r2 by:

T; = T,,,, l<i<N. (39)

Incorporating the boundary conditions, we build up
the corresponding system matrix C’E PP”). Then, in
order to find a solution to the semi-implicit scheme
[equation (34)], in each time step j, we have to solve
the linear system:
QT’+ 1 = D,
(40)
r
I
r1 R with a vector D’ E RN.
Fig. 6. Half the cross section of a cylindrical steel bar with Defining j, by
height H and radius R.
j, 6t = t,, (41)
and assuming that no martensite will be formed
and the surrounding air, respectively. We will
during the first j, steps we end up with the following
approximate the solution to equations (26) and (29)
algorithm:
by using a semi-implicit Crank-Nicholson scheme.
Defining: Initialize
70 = A,
t,=j.St, ,j=O ,..., N,, (30) p” = 0
ra = p cb, p = 0, . . , N,, (31) m” = 0.

z,. = v &, v = 0,. . , N,, (32) Forj=Otojr-220

build up 0, D
we obtain a lattice on Q x (0, fE) with the mesh sizes: solve c/T/ + I = D’
ml+’ = 0,
st+, a,=$, sz=g. (33) p’+ ’= 0.
, I
Forj=j,- 1 do
Let T,,,, be an approximation of T(r !A,z “3 t.)
, >then, for
build up C,, D
O<j<N,,O<p<N,,O<~<N~weconsiderthe
following Crank-Nicholson scheme: solve C/T’+ ’= D’
calculate initial value po,, by applying Newton’s
A(T,,,,) TN.,,+;, T,.,, method to:

with
,6t
+---- 1 1

e, = T,+1.r.l - 2T4.1+ G- 1.s.1 + 1 &+~,v,j- TN- 1.u.1 2 em,P) ’

(6rj2 dr 26r /+I -
P - Pa
mJ+’ = 0.
+ Tw + I,/- ZT,,,, + Tw - ,,,
(35)
(6z)2 Forj=j, to N,- 1 do
Linearizing the radiation condition on r2, we build up CJ, D
incorporate the boundary conditions in the usual solve CJ71+’= D
way. Owing to the linearization: calculate ml+‘, p”‘.
(36) The most time-consuming part of the algorithm is the
T(r, z, t) = T(0, z, t) + rT,(O, z, t)
numerical approximation of equation (17). In each
for r-c< 1 and arbitrary z and t, using the boundary time step a new value for p occurs in the integrand.
4382 HijMBERG: A NUMERICAL SIMULATION OF THE JOMINY END-QUENCH TEST

Hence, the integrand has to be evaluated completely 3.2. Physical parameters

in each time step, whereby the computing effort to
approximate the integral increases quadratically in
time.
To avoid a further increase in computing time, for
the calculation of ($+I, mi+‘) an explicit single-step
method was used.
The heat conductivity at 0°C has been calculated
according to Simidu’s formula (cf. [lo]):
k0 = 1.16 . (60.0 - 8.7C - 14.4Mn - 29.0%) -&,
(42)

(a)
EE 2

7.2oa

6.400

5.6OC

4.8OC

T 4.OOC

3.2OC

2.4OC

1.6OC

0.8OC

-0.400 -0.150 0.100 0.350 0.600 0.850 1.100 1.350 1.600 EE 0

log t
(b)
EE2

7.200

6.400 -

5.600 7

4.800 -

T 4.000 -

3.200 -

2.400 -_-_______ -_______

1.600

0.800

0
-0.500 -0.200 -0.100 -0.400 -0.700 1.ooo 1.300 1.600 l.!Wb 2.200 EEC

log t

Fig. 7. Numerical simulation of a CCT-diagram for the steel C 1080: (a) using the Koistinen-Marburger
formula, (b) using a rate law to describe the martensite fraction.
 HOMBERG: A NUMERICAL SIMULATION OF THE JOMINY END-QUENCH TEST 4383

(b)

-
E Q Hardenability 80

E
.: 60

20

--I
0 8 16 24 32 40 0 8 16 24 32 40
Distance from quenched end (l/16 in. units) Distance from quenched end (1116 in. units)
Fig. 8. (a) Hardenability curve for the steel C 1080 (from [2]), (b) numerically calculated martensite
fraction plotted against the distance from the quenched end.

where C, Mn, Si are the volume fractions of carbon, Finally, the temperature thresholds A,, M: can be
manganese and silicon for the respective steel. The drawn from the respective IT diagram.
temperature-dependent correction values for k and
the data for specific heat c and density p have been
4. NUMERICAL SIMULATIONS
taken from tables in [lo].
For the latent heats we take the values from [14]: 4.1. Results for the steel C 1080
First, we applied our numerical scheme to the
L, = 17.0 ;, L, = 84.0 f. (43) eutectoid carbon steel C 1080 from [2] (see Fig. 2).
Figure 7 shows the CCT diagram. As expected, the
The heat transfer coefficient has been chosen as in curves are moved to later time and lower tempera-
[ll]: ture. The bucklings of the cooling curves between the
transformation lines indicate the release of latent heat
y(T) = -0.167 x lo4 + 0.108 x lo3 x T- 0.977 during the formation of pearlite. In Fig. 7(a), we used
the Koistinen-Marburger formula [cf. equation (14)]
x 10-l. T2sK. (4) to describe the evolution of the martensite fraction.
Instead of intersecting the dotted M, line only once,

u 700
0
.s 600
e
2 500
g
al 400
E

g 300

1 10 104 103 102 10’

Sekunden
1 10 100 1000
- Zeit - Minuten
Fig. 9. IT diagram for the steel C 100 W 1 (from [23]).
4384 HtjMBERG: A NUMERICAL SIMULATION OF THE JOMINY END-QUENCH TEST

EE 2

1 \i --... \ -\‘--.

1.600 -

0.800 -

0~
-0.500 -0.200 -0.100 -0.400 -0.700 1.000 1.300 1.600 1.900 2.200 E@c
log I

Fig. 10. Numerical simulation of a CCT diagram for the steel C 100 W 1.

the cooling curves go up again. To prevent repeated
oscillations we even had to cut the latent heat Lp in
halves.
To overcome this unphysical behaviour, we
replaced the Koistinen-Marburger formula with a
rate law, which takes care of the irreversibility of the
phase change [cf. equations (15a) and (b)]. The
resulting CCT curve is depicted in Fig. 7(b). Using
the original value for L,, the cooling curves intersect
the A& line only once without performing unreason-
able heating-up effects.
Finally, Fig. 8 shows the hardenability curve for C
1080 side by side with a diagram in which the
martensite fraction is plotted against the distance
from the quenched end. Obviously, pearlite also has
a certain hardness, so one can only expect that both
curves coincide for small distances from the quenched
end, which is almost the case.
4.2. Results for the steel C 100 W I
In a second simulation, we applied our scheme to
the steel C 100 W 1 from [23]. Although this steel has
a carbon content of 1.O%, during continuous cooling
it only performs the eutectoid transformation. Thus,
the application of our model is justified.
Figure 9 shows the IT diagram for this steel. The

70 l-
I
100 _.......................,.......
.............j. f_.__._ ..,..

60

80 . .._.._.
._........ I....._..........
,...f_.__. ..~ /.....................

o 60 __....................
.,.....
_..____j.____ .!-- F. i .,. .,...,.
.z
a
Y
I 40 _ ..~.................
/ ......I .., i .,...,.,.
E:

20 _....................
f .j ../

10 I A
0 10 20 30 40 50 60 0 1 2 3 4 5
Abstand von der abgcschrecktcn Stirnflltchc in mm Distance from quenched end (cm)

Fig. 11. Hardenability curve for the steel C 100 W 1 (from [23]), left; numerically calculated martensite
fraction plotted against the distance from the quenched end, right.
 HOMBERG: A NUMERICAL SIMULATION OF THE JOMINY END-QUENCH TEST 4385

pearlite transformation starts much earlier than in the REFERENCES

case of the steel C 1080. Figure 10 depicts the
I. P. K. Agarwal and J. K. Brimacombe. M~tu/l. Trans.
numerically simulated CCT diagram for C 100 W 1
B, 12, 121 (1981).
using the rate law in equations (15a) and (b). Here, 2. American Society for Metals, Atlas of I.sothermal
the influence of the latent heat of the pearlitic Transformation and Cooling Transformation Diugrams.
transformation is more distinct than in Fig. 7. Ohio (1977).
3. D. Andreucci, A. Fasano and M. Primicerio, On a
Finally, Fig. 11 shows that for C 100 W I the
Mathematical Model,for the Crystalli-_ation of Polymers
numerically calculated martensite fraction plotted Proc. ICIAM 1991 (edited by R. E. O’Malley),
against the distance from the quenched end is beneath pp. 99-l 18. SIAM, Philadelphia (1992).
the measured hardenability curve from [23]. 4. S. H. Avner, Introduction to Ph~stt~al Metallurgy.
McGraw-Hill, Tokyo (1974).
5. M. Avrami, J. Chem. Phys. 8, 8 12 (1940).
5. CONCLUSIONS 6. J. Bergmann and T. Seeger, Z. Wetkstoffiech. 8, 89
(1977).
We have presented a model for the diffusive 7. G. Buza, H. P. Hougardy and M. Gergely, SteeI Re.s.
57, 650 (1986).
austenite-pearlite coupled with the non-diffusive
8. J. W. Cahn, Acia metall. 4, 512 (1956).
austeniteemartensite phase transition. From a math- 9. J. W. Christian, The Theory of Transformations in
ematical point of view, it can be proved that the Metals and Allow. Pergamon Press, Oxford (1975).
model is well posed. To investigate the validity of the 10. Energie- und Betriebswirtschaftsstelle des Vereins
model, we have used it to simulate the Jominy Deutscher Eisenhiittenleute, Anhaltszahlen .fir Wiirme-
btirtschaft in Eisenhiittenwerken. Verlag Stahleisen
end-quench test for two different steels.
mbH. Dusseldorf (1986).
The numerical simulations produce qualitatively 11. P. Graja, H. Miiller and E. Macherauch. Eigenspan-
reasonable results. Nevertheless, in both cases the nungsmessung an Stirnabschreckproben. Eigenspannun-
simulated hardenability curves are beneath the gen. Deutsche Gesellschaft fur Metallkunsde e.V..
Oberursel (1983).
measured ones. The reason is that the coefficients p,
12. W. J. Hayes. M.Sc. Thesis, Oxford (198.5).
c and k have been discretized explicitly. The 13. E. B. Hawbolt, B. Chau and J. K. Brimacombe. Metall.
expression k( T)/p( T)c( 7’) increases with decreasing Trans. A 14, 1803 (1983).
temperature T. Hence, owing to high temperature 14. F. Hengerer, B. Strassle and P. Bremi. Stahl u. Eisen 89,
gradients at the quenched end this explicit discretiza- 641 (1969).
15. D. Hornberg. IMA J. appl. Math. 54, 31 (1905).
tion leads to an underestimation of the heat diffusion,
16. D. Hornberg, Math. Meth. Appl. Sci. to be published.
which slows down the cooling artificially. 17. E. Hornbogen and B. Skrotzki. Steel Res. 63. 348
Thus, to improve the algorithm, one should replace (1992).
the linear equation (40) with its non-linear version: 18. H. P. Hougardy, Darstellung der (im~~andlungen fur
technische Amcendungen und Mtiglichkeiten ihrer Beei-
c’i IT”’ = D’ nflussung, Werkstoffiunde Stahl. Bd. I, pp. 198-231.
(40’)
Grundlagen, Springer Verlag, Berlin (1984).
which can be solved, e.g. by Newton’s method. 19. H. P. Hougardy and K. Yamazaki. Steel Re.s 57, 466
In order to make our model utilizable for practical (1986).
20. International Organization for Standardization. Inter-
applications, it first has to be extended to a broader
national Standard 642, Steel-Hardenability test by end
class of steels. Therefore, the formation of ferrite and quenching. Jominy test (1979).
bainite has to be incorporated. These phase 21. W. A. Johnson and R. F. Mehl, Trans. Amer. Inst. min.
transitions can be modelled similarly to the growth of metallurg. Eng., Iron Steel Dir. 135, 416 (1939).
pearlite. A further interesting line of research is to 22. D. P. Koistinen and R. E. Marburger. .Actcrnwtall. 7,
59 (1959).
incorporate the reverse transformation to austenite,
23. Max-Planck-Institut fiir Eisenforschung und der Werk-
including hysteresis effects. Then one would be able stoffausschuss des Vereins Deutscher Eisenhittenleute.
to simulate complete heat treatment cycles, giving rise Atlas z.tr Wtirmebehandlung der Stiihle, Teil I + II.
to many practical applications. Verlag Stahleisen mbH, Dusseldorf (1961).
24. E. Scheil. Arch. Eisenhiittenv~es. 12. 565 (1935)
25. C. Verdi and A. Visintin. Acta metall. 35(li), 2711
Ackno,~,ledgemmt-The author is grateful to the reviewer (1987).
for pointing out reference [ll]. Using the heat transfer 26. A. Visintin, IMA J. appl. M&h. 39, 143 (1987).
coefficient function stated there greatly helped to improve 27. A. Visintin, in Boll. U.M.I. Analisi Funzionale e
the numerical results. App/ica:ioni. Serie VI V-C 293 ~3I I (1986).