Applied Thermal Engineering 29 (2009) 29492953

Contents lists available at ScienceDirect

Applied Thermal Engineering

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

Development and application of online Stelmor Controlled Cooling System

Yu Wan-Hua a,*, Chen Shao-Hui b, Kuang Yong-Hai b, Cao Kai-Chao b
a
b

School of Materials Science and Engineering, University of Science and Technology Beijing, 30 XueYuan Road, Beijing 100083, PR China
Jiang-Su Sha-Steel Group Co., Ltd., Zhang-Jia-Gang City 215625, PR China

a r t i c l e

i n f o

Article history:
Received 3 March 2008
Accepted 3 March 2009
Available online 25 March 2009
Keywords:
Controlled cooling
Modeling
Stelmor
Rod production
Quality prediction

a b s t r a c t
An online Stelmor Controlled Cooling System (SCCS) has been developed successfully for the Stelmor production line, which can communicate with the material ow management system and Program Logic
Control System (PLCs) automatically through local network. This online model adopts Implicit Finite Difference Time Domain (FDTD) method to calculate temperature evolution and phase transformation during the production process and predicts nal properties. As Continuous Cooling Temperature (CCT) curves
of various steels can be coupled in the model, it can predict the latent heat rise and range of phase transformation for various steels, which can provide direct guidance for new steel development and optimization of present Stelmor cooling process. This unique online system has been installed in three
Stelmor production lines at present with good results.
Crown Copyright 2009 Published by Elsevier Ltd. All rights reserved.

1. Introduction
With increasing tough competition in the steel industry, how
to develop new steel products and stabilize quality of present
products becomes the major concern for steel producers. Pushing
the mechanical properties of rod wire closer to its technical limits, the demand on more reliable predictive control technique for
the cooling process in Stelmor production line increases
continuously.
Stelmor is the most popular controlled cooling process to produce the steel wire due to its fast production speed and homogeneous mechanical properties along the length of wire coil. In
Stelmor process as shown in Fig. 1, a rod wire with temperature
above 1000 C coming from the nishing mill quickly passes
through several water tanks to the laying head at a specic temperature to form into loops, depositing on to a conveyor in an overlapping pattern, the specic cooling rate is achieved by opening of
a series of fans below.
The nal mechanical properties depend mainly on the chemical
composition and the cooling rate before the phase transformation
for high carbon steel [1,2,6]. As the cooling rate and phase transformation cannot be observed directly during production, it is urgently required to develop an online model to predict the nal
mechanical properties and phase transformation. Although there
are several research reports in this eld [36], the online quality
prediction model for Stelmor process has not found reported yet.
* Corresponding author. Tel.: +86 010 62332572, +86 13381186768; fax: +86 10
82381466.
E-mail addresses: [email protected], [email protected] (W.-H.
Yu).

Adopting Implicit Finite Difference Time Domain (FDTD) method,

an online controlled cooling model was developed and installed
to monitor the real production process. This paper focuses on
introduction of its basic theory and control method of the online
system, which is helpful for stabilization of the product quality.
At present, this online model SCCS has been installed in three Stelmor production lines with satisfactory performance.

2. Mathematical model
2.1. Thermal model
In the Stelmor production line, axial heat conduction can be ignored because of an innitely long steel rod moving at high speed,
the model can be formulated to solve 1D heat conduction based on
following assumptions: (1) radial symmetry; (2) uniform circular
cross-section; and (3) uniform initial temperature, which is briey
close to reality.
Basic equation to solve the heat ow within the rod is
following:



@
@T
k @T
@T

k
gT qC p
@r
@r
r @r
@t

Note that g(T) is the volumetric rate of heat generation within

the rod due to phase transformation, q material density, Cp the
heat capacity and k the thermal conductivity.
In order to reduce loss of material message during calculation,
real experimental data q, k, and Cp can be input directly in the
interface of model as shown in Figs. 2 and 3, compared with traditional regression method.

1359-4311/$ - see front matter Crown Copyright 2009 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.applthermaleng.2009.03.012

2950

W.-H. Yu et al. / Applied Thermal Engineering 29 (2009) 29492953

Fig. 1. Layout of Stelmor machine.

Cp/J/kg C

where t is time in second, r0 the radius of rod wire, h the heat transfer coefcient. T r0 rod surface temperature, Ta air or water temperature surround.
When the online model is running, Tin is the measured temperature coming from the pyrometer installed after the nishing
mill.
The model divides the Stelmor line into several stages, each
with its constant heat transfer coefcient h value, which can be
self-adapted to match real measurements from several pyrometers
installed in the production line. In one Stelmor production line in
Sha-Steel company, there are 14 fan machines put below in sequence to control the cooling rate of steel rod by opening its fan
volume. A constant h is assigned to each fan machine controlled
stage, which can be self-adapted according to the following selfdeveloped equation:

o
Temperature/ C

Fig. 2. The steels specic heat change with temperature.

hnew hold  T C  T air =T M  T air

70

K /J/m/sec/ o C

60
50
40
30
20
10
0

200

400

600

800

1000

1200

Temperature/ oC
Fig. 3. The steels thermal conductivity change with temperature.

The FDTD CrankNicolson method has been adopted to solve

above equation as it is fast and unconditionally stable, which can
meet the requirement of online modeling.
In order to maintain balance between speed and accuracy of the
model, 20 nodes has been selected along radial direction after trial
and error. In general, large number of nodes is helpful to guarantee
accuracy of the model, but it will cause the slow speed of the model, which is unacceptable for the online model, as in real production, one rod wire can pass the laying head in no more than 2 s.
The following boundary conditions have been applied:

where hold is the original heat transfer coefcient, TC the model predicted temperature at one pyrometer position on the fan cooling
production, TM the measured temperature from the specied
pyrometer, and Tair the air temperature.
The computer model was checked by comparison between the
FDTD model and commercial FEM model FEMLAB under the exact
same conditions, Fig. 4 shows the exact same results from both
models at different condition and proves the accuracy of the FDTD
model.
As an online model, the SCCS model needs to communicate with
the material ow management system and PLC automatically
through local network continuously. The model needs to input
two groups of data: one is the real temperature from online
pyrometers, another is the basic wire rod information. Eight
pyrometers have been installed in the production line to provide
real temperature data for the model to calculate and compare.

 At the centerline

t > 0;

r 0;

k

@T
0
@r

@T
hT r0  T a
@r

 At the rod surface

t > 0;

r r0 ;

k

where the initial condition is:

t 0;

0 6 r 6 r0 ;

T T in

Fig. 4. Comparison of FDTD and FEM modeling result.

2951

W.-H. Yu et al. / Applied Thermal Engineering 29 (2009) 29492953

These temperature data is rst detected by online pyrometers, directly passed to specic PLC, then saved to Oracle database in the
database sever by the model designed by Intouch software, nally
the model can read these data from the database at every 300 ms.
another group of data that the model also needs to know is basic
information of slab, such as its composition and steel serial number
and rod diameter. These data are delivered to the specic database
once slab comes out from the reheating furnace.
2.2. Phase transformation model
How to model the phase transformation process is critical to the
success of the model as the various cooling rates during phase
transformation can determine the nal microstructure. The model
solved this problem by coupling of heat transfer analysis with
phase transformation and microstructure changes, all unied in
what is already called Microstructural Engineering.
Phase transformation process is described by following Avrami
equation [7]:



X 1  exp bTtnT

where X is the transformation fraction, b(T), n(T) are parameters

that vary with temperature, steel composition and austenite grain
size. User can input these parameters in the model directly based
on their experimental data.
One of the important parameters that need to determine is
starting temperature of austenite to pearlite transformation. The
model can provide user the interface to input CCT experimental
data of specic steels, then model can internally calculate the temperature according to the following relation [8]:

 m
dT
T AV T A1  a
dt

where a and m can be regressively obtained from CCT gure. TA1 is

the holding temperature during CCT experiment which can be input
in the model directly.
The additivity principle was adopted to calculate the heat and
amount of phase transformation. During continuous cooling stage,
at every timetemperature stage, previous phase change amount
can be transferred to the corresponding virtual time [7]:

2
t 4
0

ln 1X1i

j1

bT ij

3 1 
5

Ti
j

h
i
i
1  exp bT ij t0 DtnT j

During this time period, amount of transformation can be described as:

DX i X ij  X ij1

Phase change heat (g(T)) can be calculated based on following

equation [6]:

gT qHT

DX
Dt

where H(T) is the volumetric rate of heat generation within the rod
due to phase transformation. For high carbon steel, it can calculated
based on following formulas:
DHc-P 70621 225:23T  0:3469T 2 6:755  105 T 3 ; T > 500 C
DHc-b 92:0 kJ=kg
DHc-M 82:6 kJ=kg

With the development of microstructural evolution models, it is

now becoming possible to predict nal rolled product microstructures with increasing condence [711]. These predicted microstructures can then be combined with the structureproperty
relationship models to calculate mechanical properties. There are
a large number of factors that contribute to the strength of a steel,
such as high carbon steel, according to the equation by Mclvor and
coworker [1], there are two major factors that affect the nal ultimate tensile strength (UTS) value: rst is the cooling rate (CR) before phase transformation and second is the chemical composition,
which can be described as:

UTS 267 log C R  293 1029%C 152%Si 210%Mn

232%Cr 5244%Nf  442%P0:5

11

It needs to mention that at the paper put forward by Mclvor and

co-worker [1], CR means the cooling rate at 700 C, which briey
refer to the cooling rate before phase transformation, as for high
carbon steel, phase transformation occurs just after 700 C. In the
CSSC model, CR is modied to refer to the cooling rate in the region
from 700 C to start temperature from austenite to pearlite transformation which can be determined from the above equation [4].
It can be seen that high cooling rate before the transformation
can increase UTS correspondingly and strong inuence of nitrogen
and phosphorus on UTS. Although the presence of elements like
phosphorus, manganese, chromium and silicon are quite useful
to increase UTS, they also bring about low ductility of steel, moreover, increase the risk of precipitation of metastable phases like
bainite and/or martensite [1]. As CR can be calculated directly by
the thermal model, chemical composition of specic slab can be
known immediately from the local material management system,
it is possible to predict the nal mechanical properties during
production.
Yield tensile strength (YTS) is calculated based on following
relation:

YTS A  UTS

12

where A is constant and selected to be 0.7 in the model.

3. Results and discussion
3.1. Stability of the online model

At step j, amount of transformation can by described as:

X ij

2.3. Property prediction model

10

It is required to make the model online one, which is a challenge

for us as no mature experience can be learnt. There are two basic
requirements for the online model: stability and accuracy. It is expected that the online model can run continuously without stop, at
the same time, the output result is accurate enough to provide production guidance. First question is how to get the related parameters directly from other sources such as several online pyrometers
and related slab parameters. After trial and error, a suitable solution has been found as shown in Fig. 5.
At present, three production lines have adopted this method to
pass data to SCCS model without any problem, SCCS model can run
in cycles of 300 ms. In each cycle, the SCCS model calculates again
the temperature model (including the phase transformation and
mechanical properties) for every rod point along the length of
the cooling section, so that at any time the temporal temperature

Pyrometer

PLC

Database sever

SCCS Station

Fig. 5. Temperature data pass process.

2952

W.-H. Yu et al. / Applied Thermal Engineering 29 (2009) 29492953

course of each rod is known and can be controlled. All predicted

data is saved in Oracle database and can be checked later.

1400

3.2. Accuracy of the thermal model

1300

3.3. Mechanical property

Mechanical property prediction is a major concern for the company, if the model is accurate enough, the company will reduce the
sample checking number which can save time and money. In ShaSteel, SWRH82B steel (C 0.81/Si 0.25/Mn 0.76/Cr 0.15) is selected
to verify the model as this steel has very strict production requirement. At rst, Eq. (11) has been coupled in the SCCS model to predict the nal tensile strength, after checking with more than 400
samples, it was found that model prediction values were relatively
higher than the real measurement on the whole as shown in Fig. 7.

UTS/ MPa

After 3 years of hard work and repeated trials, the online model
has been installed in Stelmor production lines in Sha-Steel company at Jiang-Su Province. After several version modication, the
online model can run normally. Eight online pyrometers have been
installed to feed the model the real temperature. One is installed
after the nishing mill, which provides the starting temperature
for the online model. Next one is installed at the laying head, which
checks the effect of several water boxes. Other six are installed at
the fan cooling stage with each section interval, which monitor
temperature evolution in the production line and compare with
modeling results. The online model will calculate the temperatures
at every point, then automatically adapted heat transfer coefcient
(HTC) at every section according to comparison between the real
measurement and modeling result, all these can be nished in cycles of 300 ms, which can meet speed requirement for the online
model. It is observed that calculated values of HTC and actual ones
measured in operating rod mills can be considerably different, the
online model can easily study its difference and modify its HTC value correspondingly, which is helpful for accurate prediction of
phase transformation and nal mechanical property. Fig. 6 shows
the comparison between the real measurement and modeling
result.
Fig. 6 shows the close match between the real measurement
and the modeling result with no more than 20 C difference. The
effect of phase transformation on the temperature evolution can
be seen clearly, which causes slight rise during phase change stage,
then drops gradually. The model can predict duration of phase
transformation, which provides guidance for technicians to adjust
the opening of fans. In general, it is expected that phase transformation can be carried out at the same temperature within the
short time (such as no more than 10 s for 8 mm steel wire), which
can guarantee the uniform microstructure, just like patenting
treatment.

Real Measurement

Model Prediction

1350

1250
1200
1150
1100
1050
1000

51

101

151

201

251

301

351

401

Number of SWRH82B sample

Fig. 7. Comparison between the real measurement and modeling result based on
Eq. (11).

Statistics analysis show that average value gap between both was
30 MPa, with standard deviation 37 MPa. Although both show the
similar trend on the whole, it is expected that accuracy of the model could be improved further.
In order to improve accuracy of the model further, the multiple
regression method was adopted to modify the coefcient among
chemical composition, tensile strength and cooling rate, a new
equation is put forward as shown:

UTS 764 28logC R 93%C 32%Si 48%Mn 1723%Cr

21%N 83%P0:5
13
It can be seen that frameworks of Eq. (11) and Eq. (13) are the
same, UTS is related to cooling rate and chemical composition, only
difference is coefcient of various parameters, which can be obtained from real production data. Based on Eq. (13), accuracy of
the model has been improved considerably, Statistics analysis
show that average value gap between both is 0.39 MPa, with standard deviation 16.8 MPa. It can be seen from Fig. 7 that both are in
close agreement, which proves reliance of the online model to provide the production guidance. Comparing with Eq. (11), it can be
seen that except Cr element, all other factors such as cooling rate
and C element have less signicant effect on the nal UTS. This
does not mean that Eq. (11) is not accurate, as Eq. (13) comes from
the one steel sample, which has not too much uctuation in chemical composition, this will cause regression deviation.
As for other high carbon steels with different composition,
treatment is similar as SWRH82B. At rst, Eq. (11) is adopted to
predict its UTS, compare with real measurement, then modify its
related coefcients based on the multi-regression result (see
Fig. 8).
1260
Real measurement

1240

Model prediction

UTS/MPa

1220
1200
1180
1160
1140
1120
1100
1080

51

101

151

201

251

301

351

401

Number of SWRH82B sample

Fig. 6. Comparison between the real measurement and modeling result.

Fig. 8. Comparison between the real measurement and modeling result based on
Eq. (13).

W.-H. Yu et al. / Applied Thermal Engineering 29 (2009) 29492953

3.4. Advantage of the online model in the production

It needs to point out that variation of UTS during mass production is inevitable as many factors such as steel composition
and cooling rate cannot maintain unchanged during whole procession. Reasonable online prediction can help to alert workers
to pay attention to some key factors such as temperatures at
various point and cooling rates. At present, the online SCCS
model has installed in one Stelmor production for more than
one year, which shows great advantage of model during production. Before installation of CSSC model, pre-xed policy
was adopted for every steel during production, workers couldnt
directly observe the phase transformation range and temperatures at various point, sometime, phase transformation could
shift outside the controlled area, which could cause serious
quality problem, such as nal microstructure to be martensite
or coarsen pearlite. Workers can do nothing only after nal
checking. After installation of the model, workers can directly
observe the production process from user-friendly interface of
the model, adjust the opening volume of various fans according
to requirement, control the temperatures at various points in
the reasonable range, and make phase transformation occur in
the exact area. According to statistics from this production line,
rate of nal steel product with ne pearlite with grade 2.5 or
below has been increased from 87.93% to 93.54% due to installation of the CSSC model, at the same time, cases of quality
argument has greatly reduced to give users enough condence
in its nal product.
4. Conclusions
An unique online quality prediction system for the Stelmor controlled cooling line was developed, following conclusions can be
obtained based on the working experience:
1. The comprehensive model the for Stelmor controlled cooling
line consists three parts: the thermal model is obtained by
FDTD method, phase change model is obtained by solving
Avrami equation based on the experimental CCT data, and
mechanical property model is physical model based on
regressing the production data, three parts are coupled
internally.

2953

2. This online model is open to get the production information

during production, which can communicate with the material
ow management system and Program Logic Control System
(PLCs) automatically through local network.
3. No model is perfect and denitely accurate, so self-adapted
function was designed for this online model to adjust the predicted temperature along the production line based on online
pyrometers and nal properties based on nal checking. Present
results prove that it is possible to predict the nal mechanical
properties with the help of this online model with standard
deviation 16.8 MPa, which meet the requirement.
After repeated modication, this online model has been put into
production for more than 2 years, which can greatly reduce the labor work and improve the stability of product quality.
References
[1] S. Jaiswal, I.D. Mclvor, Microalloyed high carbon steel rod, Ironmaking and
Steelmaking 16 (1) (1989) 4954.
[2] I.D. Mclvor, Microalloyed low carbon steel rod, Ironmaking and Steelmaking 16
(1) (1989) 5562.
[3] Prakash K. Agarwal, J.K. Brimacombe, Mathematical model of heat ow and
austenitepearlite transformation in eutectoid carbon steel rods for wire,
Metallurgical Transactions B 12B (1981) 121133.
[4] R.D. Morales, A.G. Lopz, I.M. Olivares, Mathematical simulation of Stelmor
process, Ironmaking and Steelmaking 18 (2) (1991) 128138.
[5] R.D. Morales, A.G. Lopz, I.M. Olivares, Heat transfer analysis during water
spray cooling of steel rods, ISIJ International 30 (1) (1990) 4857.
[6] Ettore Anelli, Application of mathematical modelling to hot rolling and
controlled cooling of wire rods and bars, ISIJ International 32 (3) (1992)
440449.
[7] E.B. Hawbolt, B. Chau, J.K. Brimacombe, Kinetics of austeniteferrite and
austenitepearlite transformations in a 1025 carbon steel, Metallurgical
Transactions A 16A (1985) 565578.
[8] S. Denis, S. Jstrm, A. Simon, Coupled temperature, stress, phase
transformation calculation model numerical illustration of the internal
stresses evolution during cooling of a eutectoid carbon steel cylinder,
Metallurgical Transactions A 18A (1987) 12031212.
[9] J. Iyer, J.K. Brimacombe, E.B. Hawbolt, Prediction of the structure and
mechanical properties of control-cooled eutectoid steel rods, in: Conf. on
Mechanical Working and Steel Processing, Chicago 1984, vol. XXII, ISS/AIME,
Penn, 1985, p. 47.
[10] A. Kumar, C. McCulloch, E.B. Hawbolt, I.V. Samarasekera, Modelling thermal
and microstructural evolution on runout table of hot strip mill, Materials
Science and Technology 7 (1991) 12111223.
[11] M. Militzer, E.B. Hawbolt, T.R. Meadowcroft, Microstructural model for hot
strip rolling of high-strength low-alloy steels, Metallurgical and Materials
Transactions A 31A (2000) 12471259.