Sei sulla pagina 1di 6

Practical finite element heat transfer

modelling for hot rolling of steels


A. Tudball and S. G. R. Brown*
A transient 3D finite element model is described that has been developed to capture thermal
variations of interest during hot rolling of steel. The model has been specifically developed to
enable eventual use by plant personnel themselves. A series of validation exercises and case
studies are presented to demonstrate the versatility of the code. Both existing and as yet untried
rolling strategies can be investigated in terms of the likely thermal history that will be experienced
anywhere in the strip. Data are provided from the stage when the slab first leaves the reheat
furnace all the way up to the arrival of the strip at the runout table.
Keywords: Hot rolling steel, Finite element model, Rolling strategies, Case studies

Introduction
Control of the temperature during the hot rolling of steel
strip is a major problem, since adverse changes in
temperature substantially affect the processing characteristics and strip quality at subsequent processing stages.
The objective of this project was to use finite element
analysis to investigate heat extraction and temperature
effects due to bar processing. Modelling the steel slab and
subsequent strip in this way provides a greater understanding of the temperature history and also simulates
changes in the process without expensive trials, and
ultimately would lead to a reduction in coil rejects due to
compromised temperature effects on the mill.
Heuristic prediction of the effects of altering specific
parameters was the original method of modelling a rolling
process but over the last 80 years, progress has been made
in developing and refining numerical methods for the
prediction of many effects. In more recent times, digital
computers have allowed wholesale numerical calculations
to be made very quickly, resulting in the solution of
problems using whole theorems14 that had hitherto only
been capable of solution by introducing a simplifying
regime into the calculation. Over the last 20 years or so,
many finite difference59 and finite element1023 hot
rolling process models have been developed. These range
from two dimensional finite difference investigations on
discrete process components, to large online process
control codes. In contrast to many previous models, the
model described in this paper has been specifically
developed to handle 3D transient heat transfer during
all stages from furnace drop out of slab to exit from the
finishing mill. Furthermore, the model has been constructed so that it can be used on a standalone PC
operated by plant personnel.

Engineering Doctorate Centre in Steel Technologyz/Materials Research


Centre*, School of Engineering, University of Wales, Swansea, Singleton
Park, Swansea SA2 8PP, UK
*Corresponding author, email s.g.r.brown@swan.ac.uk

2006 Institute of Materials, Minerals and Mining


Published by Maney on behalf of the Institute
Received 5 January 2005; accepted 21 June 2005
DOI 10.1179/174328105X71317

A fully three dimensional physically based simulation


of the hot rolling process that could be operated by
untrained personnel working on the line and able to run
on standard PC platforms has now been developed.

Numerical model
As with any model, assumptions are made, the main
ones associated with this model being as follows:
1. The work roll arc of contact is assumed circular,
therefore no roll flattening is implied.
2. Simplifications regarding the workpiece/work roll
contact friction assumes a constant effect through the
arc of contact.
3. The need for strain analyses is removed with no
deformation in the z direction.
4. Atmospheric conditions remain constant.
5. Water effects behave in a uniform manner with
respect to slab/strip cooling.
The mesh created for the slab is uniform, conveniently
the hexahedral elements fit neatly into the modelled
shape. During deformation, a constant section is
maintained, which shows no bulge on the material edge,
which occurs to some degree in reality. Elements are
concentrated on the surfaces and towards the ends of the
modelled shape, to enhance comparison of heat transfer
effects and to improve output detail. Elongation of the
model means that all elements extend in the y direction
during deformation.
The model comprises a three dimensional finite
element (FE) code. Linear 8 node hexahedral elements
are used (typically 2000 elements with 2541 nodes). The
main finite element heat transfer equations are given
here for completeness24
:
p T zkfT g~f f g
(1)
where
p~

rcNT NdV

Ironmaking and Steelmaking

2006

VOL

33

NO

61

Finite element heat transfer modelling for hot rolling steel

Tudball and Brown

1 FE mesh for initial slab dimensions

k~

BRBdV z

hNT NdS3

S3

ff g~ NT fQgdV
z NT fqgdS2 z
v

S2
T

hN fT0 gdS3

S3

where r is density, kg m3
c is specific heat capacity, J kg1 K1
N is the field variable interpolation model (matrix)
B is the field variable gradient model (matrix)
R is the material property matrix
h is the heat transfer coefficient, kW m2 K1
- is the adiabatic heating term, W m2
Q
q is the heat flux, W m2
T0 is the originating temperature, uC.
dV denotes volume terms
dS denotes surface terms.
Heat sinks and sources are implemented via user defined
heat transfer coefficients with the atmosphere, the rolls and
various sprays. Experimental temperature dependent heat
transfer coefficients for sprays were used.25 This is coupled
with a simplified model for adiabatic heating of the steel
during deformation, given by26
DTm ~8:69Kw ln

h1
h2

(2)

where
h1,h2 are entry and exit thicknesses of rolled material,
m
Kw is the resistance to deformation, N m2

62

Ironmaking and Steelmaking

2006

VOL

33

NO

DT5 temperature rise in material due to mechanical


work, uC.
For speed of calculation the solution of the finite
element equations was carried out using a preconditioned conjugate gradient algorithm. Stress analysis has
not been included at this stage. From experience, the
addition of three extra degrees of freedom to be solved
using a non-linear stress analysis would increase run
times by at least an order of magnitude. Similarly, the z
direction deformation has not been included, this is a
typical assumption made in flat rolling models.2729
The FE mesh is initially set up to match the slab
dimensions and temperature of the slab as it first leaves
the reheat furnace (Fig. 1). A regular mesh is used,
although node density increases from centre to surface.
A user defined velocity is assigned to the tail end of the
slab and as it moves, particular user defined events await
it as time proceeds. These events constitute particular
types of spraying procedures and rolling operations.
While treatment of thermal effects via heat transfer
coefficients (Table 1) is handled in a conventional
manner, two approximations are made to describe shape
changes of the steel during rolling.
Each roll/pass has user defined roll radii and pass
reduction in the vertical, y direction. As surface nodes
reach and pass through a roll their y coordinates are
forced to follow the profile of the roll. For top surface
nodes this will mean a decrease and for bottom surface
nodes an increase in the y coordinate. The y coordinate
of each node between these two nodes is adjusted
proportionally. Since volume is to be conserved at all
stages in the simulation, this loss in cross-section is
compensated for by an appropriate increase in slab
length in the horizontal, x direction. Thus, as material
passes through a roll its thickness in the y direction

Tudball and Brown

Finite element heat transfer modelling for hot rolling steel

2 Template FE mesh used to simulate heat transfer in the Port Talbot coilbox

decreases and its length in the x direction increases. This


also gives rise to an acceleration of nodal points
downstream of the roll. In the current model, changes
in thickness are compensated for only by an increase
in length and velocity, no deformation to the grid is
made in the width, z direction. While this is clearly an
approximation, the simulated shape changes are in
accordance with reality and the requirement for stress
analysis is removed. Care has been taken to ensure that
the time increments are small enough to ensure all roll
reduction effects are captured accurately (typically
0?01 s). The main consequences of this procedure are
heat loss to the roll while surface nodes are in contact
with the roll and an overall increase in length and
therefore surface area of the slab that significantly
affects surface heat loss due to spraying and atmospheric
transfer. Adiabatic heating is generated by determining
the effective strain occurring in any vertical section at
any given time step (equation (2)).
In the case of the Corus Mill at Port Talbot, there is in
addition the presence of an intermediate coil box to be
considered. A special routine has been written to deal
with the presence of this coilbox, which is used to

temporarily store and also reverse the strip prior to entry


into the final descaler and finishing stands. For any
given thickness and length of strip the program
calculates a template FE mesh (Fig. 2). This mesh will
vary according to the details of individual strips. The
model maps the current strip temperature distribution
onto this template, which is then used to calculate heat
transfer during holding time in the coil box. Clearly, the
boundary conditions experienced by the surface nodes of
the strip may now be very different in that many nodes
are now effectively in contact with other surface nodes,
these being wrapped around onto them in the coiling
process. Straightforward heat conduction, with heat loss
to atmosphere being dealt with by appropriate heat
transfer coefficients occurs during this phase. At the user
defined uncoiling time the template temperature distribution is mapped back onto the strip FE mesh and the
simulation continues as before.
A schematic diagram of the hot mill at Port Talbot is
shown in Fig. 3.
Three dimensional temperature information is available at all times and all positions during the above
simulations, which can either be output as colour coded

Table 1 Thermophysical constants used in the model


Property

Constant

k (thermal conductivity)
r (density)
Cp (specific heat)
h atm(heat transfer coefficient to atmosphere)
h rolls (roughing mill)
h rolls (finishing mill)
h edge rolls
H sprays(temperature dependent)

23.16z51?96exp(22?03T/1000), W m1
7854 kg m3
834 J kg1 K1
50 W m2 K1
5000 W m2 K1
20 000 W m2 K1
17 500 W m2 K1
1004000 W m2 K1

Ironmaking and Steelmaking

2006

VOL

33

NO

63

Tudball and Brown

Finite element heat transfer modelling for hot rolling steel

3 Schematic diagram of hot rolling mill at Port Talbot

4 Temperature distribution after 45 s, where cooling of


the surface by a spray/roll combination and deformation of the FE mesh can be seen

5 Comparison of measured and predicted temperatures


for Port Talbot pilot mill

contour maps (Fig. 4) or timetemperature plots


(Fig. 5).

Case studies and results


Several case studies have been used, both to validate the
model against experimentally measured temperatures
and also to simulate certain unusual events of the type
that may occur during any real manufacturing process.
A typical C/Mn steel grade was chosen, since this
material is commonly rolled in strip mills worldwide.

In the first case study the pilot mill at Llanwern was


simulated where the equipment consists of a small
furnace to reheat trial workpieces, a single stand
reversing mill for the reduction of the workpieces, and
a runout table with water curtains.
Initially, a workpiece, of dimension 406406120 mm,
was heated to approximately 1250uC. After a period of
soaking at this temperature, the workpiece was removed
from the furnace and manually descaled by hammering
the steel surface. Subsequently, it passed into the pilot
reversing mill for seven reduction passes. Prior to each
pass, a pyrometer captured the top surface temperature
of the material. The length in the rolling direction was
initially 40 mm, and as a result, temperatures from early
passes could be missed by the instrumentation. After
each pass, the mill was reversed and the workpiece rolled
further. After seven passes, the material was transferred
to a pilot runout table for further analysis.
The availability of this equipment was useful in that
no sprays were present on the pilot mill meaning the
effects of striproll heat transfer could be concentrated
on without having the extra effects of sprays. Figure 5
shows both measured and predicted temperature evolution. Surface measurements, given the limitations of
accurately measuring surface temperatures, were in good
agreement. The temperature history at the centre of the
slab was also shown, which approached surface temperatures as increasing reduction was carried out, as
would be expected.
The next simulation, shown in Fig. 6, is for a
complete hot rolling process in Port Talbot for a slab
of dimensions 8?8661?0760?234 m. Several key events
can be identified. At position A, the material enters the
first scale breaker where after an initial atmospheric heat
loss, a temperature drop arising from the descaling spray
and subsequent slab/work roll contact is seen to be
significant. The temperature on the slab surface then
recovers somewhat, while still subject to atmospheric
heat loss. The slab then enters a reversing type

A initial descaler; B roughing passes; C coilbox; D finshing scale breaker and finishing mill; E runout table
6 Temperature history for the entire hot rolling process at Port Talbot

64

Ironmaking and Steelmaking

2006

VOL

33

NO

Tudball and Brown

Finite element heat transfer modelling for hot rolling steel

7 More detailed view of temperature history for strip


passing through finishing stands

8 Simulated effect of dummying mill stand F6 (second


mill in finishing train)

roughing mill with descaling sprays and edge mills. A


typical rolling pattern consists of five passes with
descaling sprays on passes 1, 3, and 5, though a seven
pass pattern is not uncommon. The model accommodates the reversing aspect of this mill by reversing the
finite element mesh. During the roughing mill stage, it is
worth noting that the centre temperature remains
constant up to the third or fourth pass, where it begins
to drop. This is due first, to the vast temperature losses
at the slab surface, forcing heat recovery from the slab
core, and second, the significant reduction and therefore
gauge of the entire slab.
After five roughing passes, the transfer bar passes into
the coilbox, which is used to reduce space requirements
in the mill, alongside the desirable effect of thermally
homogenising the workpiece. The surface and interior
nodes used to generate Fig. 6 show little or no
temperature change during the 20 s in the coilbox due
to the fact that they are both located internally within
the coilbox template mesh, i.e. effectively shielded from
atmospheric heat loss. From the coilbox, the transfer
bar then passes into the finishing scale breaker (FSB),
which consists of four high pressure water descaling
headers. The thermal effect at this stage is noticeable
before the material passes into the seven stand finishing
mill.
Figure 7 shows an expanded view of the same results
for the finishing stands where temperature drops can be
seen for the initial stands in the mill train. Sharp drops
are attributable to the mill stands with the minor drops
occurring between them indicating the effects of interstand cooling sprays. Adiabatic heating effects can also
be observed on the centreline curve, however, the rate of
heat loss effectively outweighs any adiabatic heating,
bearing in mind that the gauge of the material is now
approaching its final thickness. Later stands appear to
have less effect, largely due to the speed with which the
strip passes. As in real life, the material speed increases
in the later stands and reduces the chance for contact
heat transfer between workpiece and rolls.
Finally, Fig. 8 shows a simulation where one of the
finishing mill stands was dummied. In this case the mill
under investigation was the second in the finishing mill
train. A process engineering issue could necessitate the
need for a mill to be dummied. That is, the mill is still
rolling material but no reduction occurs and the mill is
not loaded, the other mills in the train performing the
reduction. Results from this simulation indicate higher
heat loss occurs with increased reduction on the more

heavily loaded mills, and small amounts of heat loss are


achieved on the dummy mill. However, the compensation of reductions has outweighed the temperature losses
previously achieved on the dummied mill and the
resulting final strip temperature is not significantly
altered.
It can be seen from Figs 5 and 6 that there is a close
correlation between the temperatures provided by the
model when compared with available industrial scale
measured data.

Discussion
The model provides satisfactory results and graphical
output allows some degree of validation. Three dimensional images of thermal maps of the slab can provide
enhanced understanding. Nodal temperatures can be
plotted for any point in the modelled slab. Presently, run
times are between 1K and 2 h when running on a 1 GHz
speed processor. In particular, Fig. 6 shows some
discrepancy between predicted and measured temperature results although all predictions are within standard
error bars. Such a fit is considered acceptable at present,
although it is anticipated that the model would be tuned
to a greater extent when applied on plant. The current
results are used merely to show that the relatively simple
model can provide relatively accurate prediction (i.e. all
within less than 10%).
Heat transfer coefficients were surveyed from previous
works and trialled in the programme. Pilot or laboratory
mill studies27 showed that values could range from 9?6
to 20?76 kW m2 K1 depending on workpiece size,
percentage reduction, roll diameter, and time in the roll
bite. Values selected for this study ranged from 2?5 to
10 kW m2 K1, which is comparable to Harding
(1976)30 who used 2.055 kW m2 K1, and Bryant and
Chiu (1982)31 who employed 7 kW m2 K1. A good
comparison was achieved using the selected heat transfer
coefficients when compared to actual pilot mill temperature readings.
A sensitivity analysis on the effects of heat transfer
coefficient between the spray and the workpiece showed
that a 10% change in heat transfer coefficient induced
a maximum predicted temperature change of only
10uC. Also, a 10% change in heat transfer coefficient to either atmosphere or roll induced a maximum
predicted temperature change of only 20uC.
While simulating the overall industrial mill configuration, it was found necessary to differentiate the roll
contact heat transfer coefficients between the roughing

Ironmaking and Steelmaking

2006

VOL

33

NO

65

Tudball and Brown

Finite element heat transfer modelling for hot rolling steel

mill section and the finishing mill section. This was


assumed to be satisfactory as the roughing mill work
rolls are the same for each pass, and also the finishing
mill work rolls had similar diameters and mill loadings.
The heat transfer coefficient chosen for the roughing
mill was 5?5 kW m2 K1 and for the finishing mill
11?5 kW m2 K1. This is comparable with existing
published works although a wide range of values has
been reported.13 Some published heat transfer coefficients are set vastly higher than the values used in
this study; in the finishing mill, a range of 75
88 kW m2 K1 has been used for one particular model,
another with a range of 1837?6 kW m2 K1.13 This is
not uncommon when comparing different process
models.
It has also been found that it is possible to apply this
model to the rolling of other metals, such as aluminium
by changing thermophysical constants, workpiece physical dimensions, and rolling configurations.

Acknowledgements
The authors would like to thank Corus Strip Products
UK for financial assistance and the sourcing of
numerous mill results. This work was carried out as
part of the EPSRC EngD scheme in the Materials
Research Centre, School of Engineering, University of
Wales Swansea.

References
1.
2.
3.
4.
5.
6.
7.
8.
9.

66

E. Orowan: Proc. Inst. Mech. Eng., 1943, 150, 140167.


D. R. Bland and H. Ford: Proc. Inst. Mech. Eng., 1943, 159, 6877.
R. B. Sims: Proc. Inst. Mech. Eng., 1954, 168, 191200.
J. M. Alexander: Proc. R. Soc., 1972, 326, 535554.
C. Devadas and I. V. Samarasekera: Ironmaking Steelmaking, 1986,
13, 311321.
W. C. Chen, I. V. Samarasekera, A. Kumar and E. B. Hawbolt:
Ironmaking Steelmaking, 1993, 20, 113125.
F. Muccardi: Proc. 35th MWSP Conf., Pittsburgh, PA, USA,
October 1993, ISS, 289295.
R. Colas: Modelling Simul. Mater. Sci. Eng., 1995, 3, 437453.
A. Purcell, N. Pokutylowicz and S. Yue: Proc. 39th MWSP Conf.,
Indianapolis, IN, USA, October 1997, ISS, 791799.

Ironmaking and Steelmaking

2006

VOL

33

NO

10. F. Yamada, K. T. Sekiguchi, M. Tsugeno, Y. Anbe, Y. Andoh,


C. Forse, M. Guernier and T. Coleman: 24th Ann. Meeting of the
IEEE Industry Applications Society IAS, San Diego, CA, USA,
October 1989, IEEE, 14451451.
11. C. Devadas, I. V. Samarasekera and E. B. Hawbolt: Metall. Trans.
A (Phys. Metall. Mater. Sci.), 1991, 22A, 307319.
12. A. Kumar, I. V. Samarasekera and E. B. Hawbolt: J. Mater.
Process. Technol., 1992, 30, 91114.
13. M. Pietrzyk and J. G. Lenard: 35th MWSP Conf. Proc., Pittsburgh,
PA, USA, October 1993, ISS, 141146.
14. R. Guo: Proc 1995 Int. Mech. Eng. Congress and Exposition, San
Francisco, USA, November 1995, ASME, 7989.
15. C. H. Huang, T. M. Ju and A. A. Tseng: Int. J. Heat Mass
Transfer, 1995, 38, 10191031.
16. Y. H. Li and C. M. Sellars: Ironmaking Steelmaking, 1996, 23, 58
61.
17. H. Dyja and P. Korczak: 3rd Int. Congress on Numerical Methods
in Engineering and Applied. Science, Merida, Venezuela, March
1996, Computational Mechanics Publishers, 359366.
18. N. Troyani and L. Montano: J. Brazilian Soc. Mech. Sci., 1999, 21,
4, 655663.
19. C. G. Sun and S. M. Hwang: ISIJ Int., 2000, 40, 794801.
20. Z.-C. Lin and C.-C. Shen: J. Mater. Process. Technol., 2001, 110, 1,
1018.
21. M. A. Cavaliere, M. B. Goldschmit and E. N. Dvorkin: Comput.
Struct., 2001, 79, 20752089.
22. M. A. Cavaliere, M. B. Goldschmit and E. N. Dvorkin: Int. J.
Numer. Meth. Eng., 2001, 52, 14111430.
23. S. Serajzadeh, H. Mirbagheri and A. Karimi Taheri: J. Mater.
Process. Technol., 2002, 125126, 8996.
24. C. S. Desai and J. F. Abel: Introduction to the finite element
method, 1st edn; 1972, New York, Van Nostrand Reinhold
Company.
25. I. C. F. Oates and I. Stewart: Determination of the heat transfer
coefficients generated by water sprays, Report Number SL/FF/
TN/3/-/94/D, British Steel Technical Internal Report, Swinden
Technology Centre, UK, 1994.
26. V. B. Ginzburg: Steel-rolling technology, 1st edn; 1989, New
York, Marcel Dekker Inc.
27. J. G. Lenard, M. Pietrzyk and C. Cser: Mathematical and physical
simulation of the properties of hot rolled products, 1st edn; 1999,
Oxford, Elsevier Science.
28. C. W. Starling: The theory and practice of flat rolling, 1st edn;
1962, London, University of London Press.
29. Z. Wusatowski: Fundamentals of rolling, 1st edn; 1969, London,
Pergamon Press.
30. R. A. Harding: Temperature and structural changes in hot rolling,
PhD dissertation, University of Sheffield, Sheffield, UK, 1976.
31. G. F. Bryant and T. S. L. Chiu: Met. Technol., 1982, 9, 478
484

Potrebbero piacerti anche