See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/272540312

Thermal fluid mathematical modelling of twin roll casting (TRC) process for
AZ31 magnesium alloy

Article  in  International Journal of Cast Metals Research · August 2013

DOI: 10.1179/1743133613Y.0000000058

CITATIONS READS
11 59

2 authors:

Amir Hadadzadeh Mary Wells

University of New Brunswick University of Waterloo
42 PUBLICATIONS   147 CITATIONS    156 PUBLICATIONS   1,406 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Optimum Design of Fatigue-Critical Automotive Components Made of Magnesium Alloys View project

Extrusion of Aluminum Alloys View project

All content following this page was uploaded by Amir Hadadzadeh on 27 October 2017.

The user has requested enhancement of the downloaded file.

 Thermal-Fluid Mathematical Modelling of the Twin Roll Casting (TRC)
Process for AZ31 Magnesium Alloy

A. Hadadzadah and M. A. Wells

Department of Mechanical and Mechatronics Engineering

University of Waterloo
Waterloo, ON, Canada

Corresponding Author: Amir Hadadzadeh

E-Mail: ahadadza@engmail.uwaterloo.ca
Phone: +1-519-888-4567 ex. 38743
Address: Mechanical and Mechatronics Engineering Department, University of
Waterloo, 200 University Avenue West, Waterloo, Ontario, Canada N2L 3G1

Second Author: Mary A. Wells

E-Mail: mawells@uwaterloo.ca
Address: Mechanical and Mechatronics Engineering Department, University of
Waterloo, 200 University Avenue West, Waterloo, Ontario, Canada N2L 3G1

1
Abstract
A two-dimensional Computational Fluid Dynamics (CFD) thermal-fluid model has been
developed and validated for twin roll casting (TRC) of AZ31 magnesium alloy using the
commercial package ANSYS® CFX®. The model was developed to represent the steady state
part of the TRC process. The thermal history of the strip was studied in terms of the temperature
gradient through the strip thickness at the exit of the caster, the sump depth and mushy zone
thickness at the centre-line for different casting speeds, final strip thickness and heat transfer
coefficient between the rolls and the surface of the strip. Moreover, the effect of these parameters
on the average solidification cooling rate and secondary dendrite arm spacing in the solidified
structure was investigated. Model validation was done by comparing the predicted and measured
exit strip surface temperature as well as the secondary dendrite arm spacing (SDAS) through the
thickness of the sheet to those measured using the Natural Resources Canada Government
Materials Laboratory, CanmetMATERIALS located in Hamilton, Ontario, Canada. Analysis of
the effect of TRC conditions on the SDAS showed that more uniform SDAS through the
thickness is obtained when casting thicker strips at higher casting speeds.

Keywords: Twin Roll Casting, AZ31 Magnesium Alloy, Thermal-Fluid Model, Secondary
Dendrite Arm Spacing.

1. Introduction
The low density of magnesium alloys makes them a desirable material for use in the
automotive industry; magnesium has a 33% lower density than aluminium and 70% lower
density relative to steel. In addition, high specific strength (   ) and stiffness ( E  ), vibration
absorption, good weldability and machinability are other attractive features of magnesium alloys
for car production1. Current applications of magnesium alloys in automotive applications are
mostly in the form of die castings. Part of the reason that sheet magnesium has seen limited
applications in automotive and other applications is the high cost of the magnesium sheet relative
to other advanced metallic alloys. This is due in large part to the manufacturing cost and
complexity to produce magnesium sheet.

2
 The traditional manufacturing route to produce magnesium sheet starts with Direct Chill
(DC) casting technology followed by several passes of hot rolling and inter-pass annealing;
typically the reduction in each rolling pass is limited to 5-20% before annealing due to the HCP
crystal structure of the magnesium and low formability2. An alternative and more cost effective
process to produce magnesium sheet is twin roll casting (TRC) which incorporates casting and
hot rolling in one process and consequently reduces the cost and energy consumption3. Twin roll
casting, known as a near-net-shape manufacturing process, is similar to rolling as the molten
material is fed into the roll bite region from the entry side and from the exit side solid sheet
material is obtained. This process has the capability of producing sheet with thicknesses between
2-10 mm4, 5 and typically experiences cooling rates in the range of 102-103 °C s-1 during
solidification5. Some of the effects of the fast cooling rate during TRC on the sheet
microstructure are an increased amount of solute in solid solution, enhanced precipitation
nucleation within the matrix, a finer distribution of precipitates and a finer grain structure6.

Although the twin roll casting process has been used in the aluminium industry for almost
60 years7, serious consideration in using this process to produce magnesium sheet was initiated
about a decade ago in 2000 2. Since this process is relatively new for magnesium sheet
production, there is a need to quantitatively understand the interaction between the process (e.g.
casting speed, set-back distance and heat transfer coefficient at roll/strip interface) and the sheet
(e.g. final thickness) as well as the necessary requirements to produce high quality sheet. The
development of knowledge-based process models of the TRC process for magnesium alloys is
one way to fulfil this goal.

During the past few years, a few models for magnesium twin roll casting process have
been developed which consider transport phenomena coupled with solidification effects. For
example, Ju et al. 8 and Hu et al. 9 modelled and analysed the effect of nozzle shape, casting
speed and set-back distance for vertical and horizontal twin roll casting of AZ31 magnesium
alloy using an FEM model in terms of flow and thermal field. The optimum casting conditions
were determined for more uniform flow field and temperature distribution. Bae et al. 10
developed a 2-D finite difference model for vertical twin roll casting of AZ91 magnesium alloy
to assess the effect of nozzle configuration and casting speed on the temperature distribution and
flow field in the roll bite region. The results were studied in terms of cooling rate and

3
solidification front position (sump depth). Zeng et al. 11 developed a CFD model to predict the
fluid flow and temperature distribution during twin roll casting of AZ31 with an asymmetric
nozzle which provided different contact lengths on the upper and lower rolls. The effects of
casting speed and exit thickness were studied and it was concluded that an asymmetric contact
zone leads to an asymmetric microstructure. Zhao et al. 4 obtained the flow field and temperature
distribution of the strip for AZ31 magnesium alloy by developing an FEM model to analyse the
effect of casting speed, strip exit thickness, heat transfer coefficient and pouring temperature.

In this study, a two-dimensional thermal-fluid model for twin roll casting of AZ31
magnesium alloy has been developed using the commercial software ANSYS® CFX®. A
constant heat transfer coefficient (HTC) at the roll/strip interface was used and the value was
chosen based on comparison between the model results to strip exit temperature and SDAS
through strip thickness measured at the Natural Resources Canada Government Materials
Laboratory, CanmetMATERIALS located in Hamilton, Ontario, Canada. The effect of casting
speed, final thickness and heat transfer coefficient at the roll/strip interface on the thermal history
and microstructure through the thickness of the strip was assessed.

2. Experimental Procedure
Fig. 1 illustrates a schematic layout of the TRC machine at CanmetMATERIALS; the
equipment consists of the following: a melting furnace, a pump and transfer tube, a headbox, a
delivery nozzle (tip), the twin roll caster stand, pinch rolls, a moving shear unit and finally
stacking unit (or coiler) 12.

The roll diameter of the TRC facility at CanmetMATERIALS is 355 mm. The rolls are
made from tool steel H13 and the facility has the ability to achieve speeds up to 6.0 m min-1. The
thickness of the cast strip can vary between 2-8 mm and the width of the strip ranges in 150-250
mm.

An electric resistance furnace is used to melt the commercial AZ31 magnesium alloy
ingots under protective gas, a mixture of SF6 and N2. A melt delivery system is used to transfer
the molten material to the headbox. The temperature of the molten material at the headbox is

4
monitored and maintained at the desired value. The gap between the two rolls is set to the desired
strip thickness and once the temperature of the nozzle reaches an appropriate value (~700°C), the
molten magnesium is fed through the nozzle to the roll bite region. Melt temperature in the
furnace, melt level and temperature in the headbox, nozzle temperature, strip surface temperature
at the exit point of the caster, casting speed, roll surface temperature and coolant water
temperature are acquired during the process. In total, eight TRC casting trials were conducted
under varying pouring temperature, casting speed and entry/exit thickness conditions. Table 1
illustrates the casting conditions conducted at CanmetMATERIALS and used for model
validation.

Fig. 1- TRC layout for magnesium alloys at CanmetMATERIALS 12.

5
 Table 1- TRC conditions conducted at CanmetMATERIALS
Trial Pouring Casting Speed Entry Thickness Exit Thickness Exit Temperature
# Temperature (°C) (m/min) (mm) (mm) (°C)
1 740 2.82 12 5 471
2 712 2.10 12 5 397
3 700 2.50 8 5 376
4 677 1.70 12 6 400
5 670 2.10 8.5 5 290
6 666 1.97 8.5 5 273
7 680 2.50 7.8 5 307
8 680 3.0 8.8 5 350

After TRC, the as-cast strips were characterized using optical metallography in terms of
microstructure through thickness and secondary dendrite arm spacing (SDAS). The samples were
chosen from the mid-width position of the strips and studied through their thickness in the
casting direction. To measure the SDAS, the primary dendrite branches were identified and the
secondary dendrite arms were considered as those developed from the primary arms13, 14. The
centre to centre distance between the neighbouring arms was then measured to determine the
SDAS. In total a minimum of 3 samples or arms were analysed for each test condition to ensure
statistically the measurements were accurate.

3. Mathematical Model Development

3.1. Geometry
As described earlier, the TRC facility consists of two counter rotating rolls between
which liquid metal is fed through a nozzle across the width of the rolls. As the liquid metal
contacts the rolls it is simultaneously cooled and pulled into the roll bite. In the roll bite the strip
continues to cool and is also deformed until at the exit it emerges as a solid strip. Fig. 2 shows
2D schematic of a twin roll caster. The figure also illustrates some of the terminology used in the
rest of the paper when discussing results. The set-back distance (𝓵1) is the distance between the
nozzle entry and the caster kissing point (the point where the least distance between two rolls
occurs). 𝓵2 is the sump depth, which is the distance between the nozzle entry and the coherency
temperature (in this case 90% fraction solid15) position along the centre-line and 𝓵3 is the mushy

6
zone thickness at the centre-line in the casting direction and determined as the distance between
the liquidus and coherency temperature positions.

The following assumptions were considered in developing the model 16:

I. The process is dominated by transport phenomena in two dimensions. In the third dimension
(across the width) there is no significant heat transfer or fluid flow as the simulation is done at
the mid-width location,

II. The process is modelled in steady state,

III. Due to symmetry only the top half of the strip and part of the top roll was modelled in the
simulation,

IV. The fluid flow is laminar,

V. The roll surface maintains a temperature of 60 °C and roll deformation is neglected.

The important physical phenomena which happen during the process and are included in
the model include:

I. Heat transfer and fluid flow in the melt sump (liquid metal, zone 1 in Fig. 2),

II. Heat transfer, fluid flow and latent heat of fusion release in the mushy zone (zone 2 in Fig. 2),

III. Heat transfer in the solid phase (zone 3 in Fig. 2), and

IV. Heat transfer from the magnesium sheet to the roll surface.

7
 Fig. 2- Schematic of the twin roll casting process; region 1 is the liquid metal, region 2 is the mushy zone
and region 3 is the solid strip, 𝓵1, 𝓵2 and 𝓵3 are set-back distance, sump depth and mushy zone thickness at centre-
line (℄), respectively. Note: The perspective of the TRC is not to scale and the nozzle size and position are
magnified with respect to the rolls.

3.2. Governing Equations

The basis of the numerical simulation of the fluid flow and heat transfer is the
conservation laws of mass, momentum and energy16. Since the process is being modelled during
steady state, time independent governing equations are considered, as shown in equations (1-3).

• Mass conservation equation:


( u j )  0 (1)
x j

• Momentum conservation equation:

  u p
( uiu j )  gi  ( i )  (2)
x j x j x j xi

• Energy conservation equation:

8
   T
( u j CPT )  (k ) (3)
x j x j x j

where the subscripts i and j show the directions, u is the velocity (in m s-1),  is the density
(in kg m-3), g the gravity vector (in m s-2),  is the dynamic viscosity (in Pa s), p is pressure (in

Pa), CP is the specific heat capacity (in J kg-1 °C-1), k is thermal conductivity (in W m-1 °C-1) and
T is temperature (in °C). To account for solidification, two important effects of this phenomenon
on the fluid flow and heat transfer need to be included; the effect of a semisolid region on the
fluid flow and the latent heat that is released during solidification. Since alloy solidification
occurs over a temperature interval, a mushy zone (mixture of solid and liquid) is formed which
will damp or inhibit fluid flow in that zone. It's assumed this mushy region acts as a porous
medium and obeys Darcy's equation 11, 17; so an additional term is added to the momentum
conservation equation and equation (2) is modified as shown in equation (4):

  u p (1  f )2
( uiu j )  gi  ( i )   C 3 l (u j  ur , j ) (4)
x j x j x j xi fl  

where C is a constant of the liquid phase between 104-107 17, 18, f l is the fraction liquid,  a

small number to avoid division by zero when the liquid fraction approaches zero and ur , j is the

roll velocity component in the j -direction (in this case it is assumed to be casting velocity in the
x -direction and 0 in other directions). By adding this source term, velocities in the mushy zone
gradually approach the corresponding component of the actual velocity, as the fraction liquid
goes to zero. The second effect of solidification; the release of latent heat, is modelled by
considering an equivalent specific heat capacity11, as shown in equation (5); which then is
implemented in the energy conservation equation.

T
H  H ref   CP dT (5)
Tref

where H is the enthalpy (in J kg-1), H ref reference enthalpy (here the latent heat of fusion, in J

kg-1), C P specific heat capacity (in J kg-1 °C-1) and Tref reference temperature (here solidus

temperature in °C).

9
3.3. Boundary Conditions
A critical aspect of the model development will be quantifying boundary conditions to
describe the twin roll casting process accurately. There are six process boundary conditions as
described below and depicted in Fig. 3:

 Inlet (region 1): the boundary condition in this region consists of the casting temperature
and either a reference pressure or velocity. The liquid velocity is calculated from the casting
speed while the mass is conserved during the process. A uniform velocity is defined at inlet. In
order to allow the flow to completely develop, the inlet is located 100 mm before the first contact
between the melt and roll (this design lets the velocity develop inside the nozzle).
 Melt container upper surface and upper part of the nozzle (region 2): this region is
assumed to be adiabatic (no heat loss) and that there is a no-slip wall condition (no relative
velocity between the fluid and the boundary).
 Roll/strip interface (region 3): In this region a no-slip rotating wall is defined. The heat
transfer between the roll and the strip is defined using a heat transfer coefficient (HTC) as shown
in equation (6). For the analysis it was assumed that HTC’s value remained constant from the
first point of contact of the liquid against the roll surface to the exit of the strip from the rolls.
The roll surface temperature was assumed to be constant at 60 °C (based on the measurements
during experimental procedure), typical of its steady state temperature.
T
k  HTC(Ts  T0 ) (6)
n

where k is the thermal conductivity (in W m-1 °C-1), T temperature (in °C), HTC heat transfer
coefficient (in W m-2 °C-1), n normal direction to the strip surface, Ts strip surface temperature

and T0 roll surface temperature (assumed to be 60 °C).

 Outlet (region 4/5): For the surface of the exit strip (region 4) radiation is neglected in
this region because of the low temperatures but a low value of HTC=12 W m-2 °C-1 11 is used to
account for heat transfer from the strip to the air. The interface is defined as free-slip wall. The
exit velocity is equal to the casting speed.

10
 Centre-line (℄): the centre-line is considered to be a symmetry boundary with no fluid
flow or heat transfer across the interface.

Fig. 3- Boundary regions on the solution domain of the twin roll casting process.

The ANSYS® CFX® commercial package was used to define the geometry, mesh, and
boundary conditions and to solve the heat transfer and fluid flow equations. The mesh used was
structured quadrilateral as illustrated in Fig. 3. The elemental size used for generating the mesh
for the strip ranged from a minimum of 0.1 mm (in the melt sump) to a maximum of 2 mm (near
the inlet and outlet) in side length.

3.4. Material Properties

Thermo-physical properties of AZ31 magnesium alloy used in the current study are
available in the literature and are shown in Table 2. In order to conduct a representative model of
the TRC process, the non-equilibrium (Scheil cooling conditions) solidus and liquidus
temperatures were used for the AZ31. Values reported by Hao et al. 19, generated by the
computational thermodynamics database JMatPro, were used. The non-equilibrium fraction solid
used for latent heat release calculation is shown in Fig. 4.

11
 Table 2- Thermo-physical properties of AZ31 magnesium alloy
Property Value/Function
Density, ρ (kg m-3) 9 1780
Latent heat of fusion, L (J kg-1) 16 340000
Specific heat, CP (J kg-1 °C-1) 20 820+(0.79×T)-((3.6×106)/(T-255)2)*
50°C 100°C 200°C 250°C 424°C 630°C 635°C 680°C
Thermal conductivity, k (W m-1 °C-1) 19
83.9 87.3 97.0 101.8 118.5 60 120 240
Solidus, Tsol (°C) 19 424
Liquidus, Tliq (°C) 16 635
Coherency Temperature TCoh (°C) 15 578**
* Temperature in Kelvin.
** At 0.9 fraction solid.

Fig. 4- Non-equilibrium fraction solid (Scheil cooling conditions) for AZ31 magnesium alloy, graph
reproduced from Reference 19.

3.5. Casting Conditions

The casting conditions used in the current study to model the twin roll casting process for
AZ31 magnesium alloy are shown in Table 3. A total of 54 simulations were run using the
various conditions shown in Table 3. The aim of the modelling work was to assess the effect of
casting speed, final thickness and heat transfer coefficient (HTC) at the roll/strip interface on the

12
thermal history and microstructure evolution of the cast strip. The final thickness affects the set-
back distance (𝓵1 in Fig. 2) which is an important parameter in controlling the twin roll casting
process. Set-back distance (SB, in mm) for the TRC process is calculated using equation (8).

SB  1  ( Rh  0.25h2 ) (8)

where R is the roll radius (in mm) and h is the reduction (the difference between entry and exit
thickness) (in mm).

The effect of casting speed, HTC and exit thickness on the thermal history in the strip
during TRC was studied. Thermal history was studied in terms of temperature gradient through
the thickness of the strip at the exit region, sump depth at the centre-line (𝓵2 in Fig. 2), mushy
zone thickness at the centre-line (𝓵3 in Fig. 2) and averaged cooling rate at different positions of
the strip. The cooling rate was then correlated to the microstructure evolution in terms of the
secondary dendrite arm spacing (SDAS).

Table 3- Casting conditions modelled in the current study

Casting Entry Exit Set-Back Heat Transfer Casting
Temperature (°C) Height Thickness Distance Coefficient (kW m-2 °C-1) Speed (m
(mm) (mm) (mm) min-1)*
4 37.5
0.5, 1.0, 1.7,
677 12 5 35.1 11, 13 & 15
2.0, 2.5 & 3.0
6 32.5
* The survey of the modelling results revealed the fact that casting speed of 0.5 m min-1 leads to failure for all final
thicknesses and heat transfer coefficients due to nozzle blocking (solidification inside the nozzle). So, for the rest of
the study casting speed of 0.5 m min-1 was excluded.

3.6. Model Validation

Model validation was performed by comparing the predicted exit strip temperature and
secondary dendrite arm spacing through the strip thickness with those measured at
CanmetMATERIALS.

One of the most important and least well known boundary conditions for the twin roll
casting process is the heat transfer coefficient (HTC) between the strip and roll surface. The HTC
can be influenced by factors such as: roll texture and roughness, thermo-physical properties of
the roll and strip material, pressure at the interface (roll pressure) and superheat of the molten
metal entering the roll gap. As a first approximation the HTC during TRC was considered to be

13
constant along the roll bite, which is an approach adopted by many other researchers4, 8-11. The
HTC can then be adjusted to so that experimental measurements of the strip temperature are
matched against model predicted ones21. In the current study, the HTC was fit against strip
surface temperature measurements made 5 cm from the exit point of the twin roll caster. Another
method to validate the heat transfer part of the TRC model was to compare model predictions of
the secondary dendrite arm spacing (SDAS) to those measured in the strip.

Cooling rates used for SDAS predictions were calculated by considering an average
solidification cooling rate at various positions through the thickness of the strip. Solidification
time from the model at a discrete location was evaluated using equation (9), which then is used
to calculate the average cooling rate using equation (10). Since 90% of solidification process is
accomplished at the coherency temperature and the dendrites are characterized at this point, the
averaged solidification cooling rate was calculated between liquidus and coherency temperature.
The results were in agreement with the measurements. According to the work done by Allen et
al. 22 the relation between cooling rate and secondary dendrite arm spacing for AZ31 magnesium
alloy follows equation (11). This empirical relation, which is valid for cooling rate ranges 10-1 to
106 °C s-1, was used in the current study to correlate the predicted cooling rates and SDAS.

x1 dx
t (9)
x0 v( x)

Tliq  TCoh
R (10)
t

  35.5R 0.31 (11)

where t is the solidification time (in s), v(x) the velocity profile in the solidification range (in m

s-1) and x0 and x1 are the solidification start and end positions (corresponding to the positions of
liquidus and coherency temperature in the temperature profile) (in m), R the averaged cooling
rate (in °C s-1), Tliq and TCoh liquidus and coherency temperature (in °C) and  secondary

dendrite arm spacing (in µm).

The model was run using various values for HTC range 10-20 kW m-2 °C-1 and then the
predicted exit temperatures were compared with the measured ones (shown in Table 1) to

14
evaluate the optimum HTC. As shown in Fig. 5, the comparison suggested the optimum value for
HTC is 11 kW m-2 °C-1 since the predicted values match the measured ones in a range of ±10%
error.

In Fig. 6 the measured SDAS through thickness versus the predicted values are shown for
two trials, using the optimum HTC value of 11 kW m-2 °C-1. For each trial two sets of results are
presented; the first set shows the discrete individual measurements of the SDAS through the
thickness and the second set shows the averaged values for SDAS at the top and bottom surfaces,
centre-line and quarter and three quarter positions. As observed, there is a fairly good agreement
(within 20%) between the measured and predicted values. This validation procedure proved the
liability and accuracy of the developed model.

Fig. 5- Comparison of predicted and measured temperature for the conditions shown in Table 1, using an
HTC value of 11 kW m-2 °C-1. The solid line shows the perfect match and two dotted lines show ±10% variance
form the correct values.

15
 (a) (b)

(c) (d)
Fig. 6- Predicted (solid line) and measured (symbol) SDAS (λ) through the strip thickness for a) trial #3 all
data, b) trial #3 averaged data, c) trial #4 all data and d) trial #4 averaged data. The dotted lines show ±20% variance
from the model predictions.

4. Results and Discussion

4.1. Microstructure Evolution

Fig. 7 shows the typical as-cast microstructure of the twin roll cast AZ31 through the
thickness of the strip. As shown, this microstructure consists of different areas, since the cooling
rate varies significantly from the strip surface to the centre-line. At the strip surface, where direct

16
contact between the molten material and roll surface occurs, the cast material experiences the
highest cooling rate during the process. Initially, a thin layer of a rapid cooled microstructure is
formed on the surface of the strip which is known as the chill zone. Below the chill zone, a
columnar dendritic zone develops in the direction of heat removal. Due to the rotation of the rolls
these columnar grains incline from the surface to the centre of the strip. By growing the
columnar zone toward the centre of the strip, the solute is rejected to the remaining liquid metal.
Due to different cooling conditions and the presence of the solute rich liquid material an
equiaxed zone is formed at the strip centre23.

Fig. 7- As-cast microstructure of twin roll cast AZ31 through thickness for trial #3.

Fig. 8 shows more details of the microstructure of the AZ31 twin roll cast strip above the
centre-line of the strip. The black arrows on the microstructure illustrate the direction of the
columnar dendrites growth.

17
 Fig. 8- Optical image of microstructure of the strip above the centre-line, the black arrows show the
columnar dendrites growth direction, the circle on lower legends shows where the sample has been chosen.

4.2. Exit Strip Temperature and Temperature Gradient through Thickness

As the molten material enters the roll bite and is in contact with the roll surface, the
temperature at the surface of the strip drops rapidly to the liquidus temperature and solidification
starts. Continuing along the arc of contact, more heat is extracted from the material as it fully
solidifies and then heat is conducted from the solid strip to the roll surface. As more heat is
extracted from the strip to the rolls, a lower temperature at the surface of the exit strip is
obtained. The model predicted exit strip surface temperature for different casting speeds and
final thicknesses is shown in Fig. 9 using an HTC= 11 kW m-2 °C-1. Increasing the casting speed
causes less time for heat transfer from the strip to the roll and as a consequence the overall
amount of heat extracted from the strip is reduced. Moreover, decreasing the final exit thickness

18
of the strip provides a longer set-back distance (equation (8)) and longer arc of contact which
then leads to more heat transfer from the cast material to the roll. As observed in Fig. 9, the
sensitivity of the process to the final thickness in terms of exit temperature is more significant at
higher casting speeds; the difference between exit temperature for final thicknesses of 4 and 6
mm at 1.0 m min-1 is 85 °C and at 3.0 m min-1 is 153 °C using an HTC=11 kW m-2 °C-1. Similar
trends were observed for HTC values of 13 and 15 kW m-2 °C-1. However, the predicted exit
temperature became more sensitive to the final thickness for higher casting speeds at higher heat
transfer coefficients; i.e. 50 °C and 171 °C difference in exit temperature for thicknesses of 4 and
6 mm at 1.0 m min-1 and 3.0 m min-1, respectively (for HTC=15 kW m-2 °C-1). Besides showing
the heat transfer behaviour of the caster, the exit strip surface temperature is a helpful parameter
in controlling the dynamic recrystallization and grain growth experienced by the twin roll cast
strip. At higher temperatures dynamic recrystallization occurs at higher fractions; follows by
grain coarsening if enough heat is retained in the strip24.

19
 Fig. 9- Model-predicted effect of casting speed and final thickness on the surface temperature of the exit
strip using an HTC=11 kW m-2 °C-1.

Fig. 10 shows the influence of heat transfer coefficient on the exit strip temperature for a
final thickness of 6 mm cast under different casting speeds. As expected, as the heat transfer
coefficient is increased the exit strip temperature is decreased for a given casting speed.
Quantitative knowledge of the thermal history is imperative as the solidification structure and
amount of deformation that occurs during TRC, as well as the final microstructure through the
strip thickness are dependent on it.

20
 Fig. 10- Model-predicted effect of casting speed and heat transfer coefficient on the surface temperature of
the exit strip for an exit strip thickness of 6 mm.

In addition to the strip surface temperature at the twin roll caster exit, the temperature
gradient at the exit region, i.e. the temperature gradient from strip surface to centre is another
important parameter. Sahai et al. 25 and Saxena et al. 26 believe a higher temperature gradient (at
the exit) induces higher stress through the thickness which increases the probability of crack
formation in the solidified strip. As illustrated in Fig. 11, a larger temperature gradient is
predicted as the casting speed increases. Moreover, there is a stronger dependency on casting
speed for thicker exit strips. As expected, thicker exit strips experience a larger temperature
gradient at the twin roll caster exit for the same casting speed. In contrast with the surface exit
temperature, the temperature gradient was not observed to be sensitive to the heat transfer
coefficient for final thicknesses of 4 and 5 mm. For the final thickness of 6mm at casting speeds
of 2.5 and 3.0 m min-1, the temperature gradient was predicted to be more sensitive to the heat
transfer coefficient for the conditions studied.

21
 Fig. 11- Model-predicted temperature gradient at the exit point of the caster for different casting speeds,
HTCs and exit strip thicknesses.

4.3. Sump Depth and Mushy Zone Thickness at the Centre-Line

The effect of casting speed and HTC on the sump depth within the TRC process for an
exit strip thickness of 4 mm is shown in Fig. 12. As the sump depth increases at higher casting
speeds and solidification front moves closer to the exit point of the caster, the strip will undergo
less plastic deformation as a consequence of the rolling process. Increasing the HTC caused the
solidification front to move further away from the exit point. Similar to the exit strip
temperature, as the casting speed increases, the process becomes more sensitive to the HTC. Fig.
13 shows the effect of final thickness on the normalised sump depth; increasing the final
thickness causes the solidification front to move closer to the exit region, significantly at high
speeds. The x-position (set-back distance) in the TRC process was normalised, such that the
entry was 0 and exit 1, by dividing the x-position of each case to its set-back distance. Moreover,

22
the mushy zone thickness at the centre-line (𝓵3 in Fig. 2) is affected by the casting parameters as
shown in Fig. 14. While raising the casting speed causes a deeper sump and a shift in the
solidification front position, simultaneously the thickness of the mushy zone increases. Also, a
deeper mushy zone was observed when casting the thicker strips. The studies done by Yun et al.
7
and Gras et al. 27 suggest the centre-line segregation formation for twin roll cast aluminium
alloys is more frequent when a deeper sump (𝓵2) occurs; more solute rich molten material is
formed in deeper mushy zones which promotes the formation of this defect.

Fig. 12- Model-predicted sump depth for a final strip thickness of 4 mm.

23
Fig. 13- Model-predicted effect of final thickness on the sump depth for HTC=11 kW m-2 °C-1.

24
 Fig. 14- Model-predicted mushy zone thickness for HTC=11 kW m-2 °C-1.

4.4. Microstructure Uniformity

The microstructure of a twin roll cast material is directly affected by the thermal history
experienced during the solidification period from the liquid phase through to final solidification.
A comprehensive understanding of the solidification cooling rate is helpful in predicting the final
microstructure of the material including features such as the Secondary Dendrite Arm Spacing
(SDAS). As described in section 3.6, the averaged solidification cooling rate can be calculated
using details of the flow field or the velocity profile at each position of the strip.

Fig. 15 shows the predicted effect of casting speed on the average cooling rate during
solidification for the AZ31 twin roll cast strip with a final thickness of 4 mm at the strip surface
and centre-line. As the casting speed increases the average cooling rate during solidification
decreases. Increasing the casting speed will cause the distance over which the solidification

25
occurs (mushy zone thickness) to increase; so, the solidification time increases and cooling rate
decreases. Fig. 16 illustrates the corresponding predicted secondary dendrite arm spacing
showing that slower casting speeds should produce more uniform final microstructures. Similar
effects are seen for higher final thicknesses.

Fig. 15- Effect of casting speed on the model-predicted cooling rate during solidification for the AZ31 twin
roll cast strip at the surface and centre-line. (Final thickness=4 mm and HTC=11 kW m-2 °C-1).

26
 Fig. 16- Model-predicted effect of casting speed on secondary dendrite arm spacing through the normalised
thickness (0 = top and 1 = bottom) for a final thickness of 4 mm and HTC=11 kW m-2 °C-1.

If the uniformity of the microstructure is defined as the difference between the SDAS on
the surface and centre-line (   centreline  surface ); a higher Δλ implies less uniformity and

vice versa. Fig. 17 shows the effect of casting speed and final thickness on the AZ31 cast strip
microstructure uniformity assuming the HTC=11 kW m-2 °C-1. As expected, increasing the
casting speed leads to the evolution of less uniform microstructure after casting. Moreover, more
uniformity is obtained for thinner strips. Fig. 18 shows the influence of heat transfer coefficient
and casting speed on the microstructure uniformity for final thickness of 4 mm; less uniformity is
accomplished as the heat transfer coefficient is increased for low casting speeds. For high casting
speeds; i.e. 2.0 m min-1 and higher, the microstructure uniformity is independent of HTC.
Therefore, the most uniform microstructure is achieved by casting the AZ31 strip at 4 mm final
thickness, casting speed of 1.0 m min-1 with HTC=11 kW m-2 °C-1.

27
Fig. 17- Model-predicted effect of casting speed and final thickness on the microstructure uniformity
through thickness for AZ31 twin roll cast strip casting with HTC=11 kW m-2 °C-1.

28
 Fig. 18- Model-predicted effect of casting speed and heat transfer coefficient on the microstructure
uniformity through the thickness of 4 mm AZ31 twin roll cast strip.

5. Conclusions
A two-dimensional Computational Fluid Dynamics (CFD) thermal-fluid model has been
developed and validated for twin roll casting (TRC) of AZ31 magnesium alloy using the
commercial package ANSYS® CFX®. The thermal history of the strip was studied in terms of the
temperature gradient through the strip thickness at the exit of the caster, the sump depth and
mushy zone thickness at the centre-line and as-cast microstructure uniformity. Model validation
was done by comparing the predicted and measured exit strip surface temperature as well as the
secondary dendrite arm spacing (SDAS) through the thickness of the sheet. The following
conclusions can be drawn from this work:

29
1) Higher casting speeds, thicker final exit gauges and lower HTC cause the strip to exit with
higher temperatures as well as increase the depth of the sump and thickness of the mushy zone.

2) Lower temperature gradient through the strip thickness is achieved by casting at lower casting
speed and reducing the final thickness. The effect of HTC on the temperature gradient is
negligible.

3) The cooling rate which occurs during solidification at the strip surface and centre-line
decreases by increasing the casting speed.

4) More uniform microstructures are obtained by casting at lower speeds, casting the strip with
lower final exit thickness and applying a lower HTC. The optimum condition in terms of
microstructure uniformity is achieved by casting the AZ31 strip to a final thickness of 4 mm
using a casting speed of 1.0 m min-1 and HTC=11 kW m-2 °C-1; however, lower casting speeds
can lead to lower productivity.

5) The process becomes more sensitive to HTC and strip thickness as the casting speed is
increased.

Acknowledgments
The authors of this work would like to appreciate the NSERC (Natural Sciences and
Engineering Research Council of Canada) Magnesium Strategic Research Network (MagNET)
for the financial support of this work and the Natural Resources Canada Government Materials
Laboratory, CanmetMATERIALS located in Hamilton, Ontario for providing the opportunity to
perform the experimental parts of the work. The assistance of Dr. M. Kozdras, Dr. A. Javaid and
Mr. G. Birsan and P. Newcombe (from CanmetMATERIALS) in processing the experimental
data is gratefully acknowledged.

30
References
1. K. U. Kainer, "Magnesium Alloys and Technology", 3; 2003, WILEY-VCH Verlag GmbH & Co. KG aA.

2. D. Liang, C. B. Cowley, "The Twin-Roll Strip Casting of Magnesium", Journal of the Minerals, Metals and
Materials Society, 2004, 56 (5), 26-28.

3. E. E. M. Luiten, K. Blok, "Stimulating R&D of Industrial Energy-Efficient Technology; The Effect of Government
Intervention on the Development of Strip Casting Technology", Energy Policy, 2003, 31, 1339-1356.

4. H. Zhao, P. Li, L. He, "Coupled Analysis of Temperature and Flow during Twin-Roll Casting of Magnesium Alloy
Strip", Journal of Materials Processing Technology, 2011, 211, 1197-1202.

5. H. S. Di, Y. L. Li, Z. L. Ning, Z. Li, X. Liu, G. D. Wang, "New Processing Technology of Twin Roll Strip Casting
of AZ31B Magnesium Strip", Materials Science Forum, 2005,488-489, 615-618.

6. H. Chen, S. B. Kang, H. Yu, H. W. Kim, G. Min, "Microstructure and mechanical properties of Mg-4.5Al-1.0Zn
alloy sheets produced by twin roll casting and sequential warm rolling", Materials Science and Engineering A,
2008, 492, 317-326.

7. M. Yun, S. Lokyer, J. D. Hunt, "Twin Roll Casting of Aluminum Alloys", Materials Science and Engineering A,
2000, 280, 116-123.

8. D. Y. Ju, H. Y. Zhao, X. D. Hu, K. Ohori, M. Tougo, "Thermal Flow Simulation on Twin Roll Casting Process
for Thin Strip Production of Magnesium Alloy", Materials Science Forum, 2005, 488-489, 439-444.

9. X. D. Hu, D. Y. Ju, H. Y. Zhao, "Thermal Flow Simulation and Concave Type Slot Nozzle Design for Twin Roll
Casting of Magnesium Alloy AZ31", Materials Science Forum, 2007,539-543, 5037-5043.

10. J. W. Bae, C. G. Kang, S. B. Kang, "Mathematical Model for the Twin Roll Type Strip Continuous Casting of
Magnesium Alloy Considering Thermal Flow Phenomena", Journal of Materials Processing Technology, 2007, 191,
251-255.

11. J. Zeng, R. Koitzsch, H. Pfeifer, B. Friedrich, "Numerical Simulation of the Twin-Roll Casting Process of
Magnesium Alloy Strip", Journal of Materials Processing Technology, 2009, 209, 2321-2328.

12. E. Essadiqi, I. H. Jung, M. A. Wells in "Advances in Wrought Magnesium Alloys-Fundamentals of Processing,

Properties and Applications", (Eds. C. Bettles, M. Barnett), 279; 2012, Woodhead Publishing Limited.

13. R. N. Grugel, "Secondary and Tertiary Dendrite Arm Spacing Relationships in Directionally Solidified Al-Si
Alloys", Journal of Material Science, 1993, 28, 677-683.

14. M. S. Turhal, T. Savaskan, "Relationships between Secondary Dendrite Arm Spacing and Mechanical
Properties of Zn-40Al-Cu Alloys", Journal of Material Science, 2003, 38, 2639 – 2646.

15. H. Hao, D. M. Maijer, M. A. Wells, A. Phillion, S. L. Cockcroft, "Modeling the Stress-Strain Behavior and Hot
Tearing during Direct Chill Casting of an AZ31 Magnesium Billet", Metallurgical and Materials Transactions A,
2010, 41A, 2067-2077.

16. A. Hadadzadeh, M. A. Wells, E. Essadiqi, "Mathematical Modelling of the Twin Roll Casting for Magnesium
Alloys - Effect of Heat Transfer Coefficient between the Roll and the Strip", Proceedings of 8th International
Conference on Magnesium Alloys and their Applications, Weimar, Germany, Oct 26-29, 2009, pp. 138-144.

17. R. I. L. Guthrie, R. P. Tavares, "Mathematical and Physical Modeling of Steel Flow and Solidification in Twin-
Roll/Horizontal Belt Thin-Strip Casting", Applied Mathematical Modeling, 1998, 22, 851-872.

31
18. A. Baserinia, H. Ng, D. C. Weckman, M. A. Wells, S. Barker, M. Gallerneault, "A Simple Model of the Mold
Boundary Condition in Direct-Chill (DC) Casting of Aluminum Alloys", Metallurgical and Materials Transactions B,
2012, 43B, 887-901.

19. H. Hao, D. M. Maijer, M. A. Wells, S. L. Cockcroft, D. Sediako, s. Hibbins, "Development and Validation of a
Thermal Model of Direct Chill Casting of AZ31 Magnesium Billets", Metallurgical and Materials Transactions A,
2004, 35A, 3843-3854.

20. Y. He, A. Javaid, E. Essadiqi, M. Shehata, "Numerical Simulation and Experimental Study of the Solidification
of a Wedge-Shaped AZ31 Mg Alloy Casting", Canadian Metallurgical Quarterly, 2009, 48, 145-156.

21. P. Bradbury, "A Mathematical Model for the Twin Roll Casting Process", PhD Thesis, Department of Materials,
University of Oxford, Oxford, UK, 1994.

22. R. V. Allen, D. R. East, T. J. Johnson, W. E. Borbidge, D. Liang, "Magnesium Alloy Sheet Produced by Twin
Roll Casting", 2001, in Magnesium Technology, TMS 2001, New Orleans, Louisiana , USA, pp. 75-79.

23. M. Aljarrah, E. Essadiqi, D. H. Kang, I. H. Jung, "Solidification Microstructure and Mechanical Properties of
Hot Rolled and Annealed Mg Sheet Produced through Twin Roll Casting Route", Materials Science Forum, Light
Metal Technology, 2011, 60, 331-334.

24. S. Das, N. S. Lim, H. W. Kim, C. G. Park, "Effect of Rolling Speed on Microstructure and Age-Hardening
Behaviour of Al-Mg-Si Alloy Produced by Twin Roll Casting Process", Materials and Design, 2011, 32, 4603-4607.

25. Y. Sahai, A. Saxena. "Modeling of Twin-Roll Thin Strip Casting of Aluminum Alloys", TMS (The Minerals,
Metals and Materials Society), 2002, in Light Metals 2002, TMS 2002, Seattle, Washington, USA, pp. 643-650.

26. A. Saxena, Y. Sahai, "Modeling of Fluid Flow and Heat Transfer in Twin-Roll Casting of Aluminum Alloys",
Materials Transactions, 2002, 43-2, 206-213.

27. C. Gras, M. Meredith, J. D. Hunt, "Micro Defects Formation During the Twin-Roll Casting of Al-Mg-Mn
Aluminum Alloys", Journal of Materials Processing Technology, 2001,167, 62-72.

32
List of Figure Captions
Fig. 19- TRC layout for magnesium alloys at CanmetMATERIALS 12.

Fig. 20- Schematic of the twin roll casting process; region 1 is the liquid metal, region 2 is the mushy zone and
region 3 is the solid strip, 𝓵1, 𝓵2 and 𝓵3 are set-back distance, sump depth and mushy zone thickness at centre-line
(℄), respectively. Note: The perspective of the TRC is not to scale and the nozzle size and position are magnified
with respect to the rolls.

Fig. 21- Boundary regions on the solution domain of the twin roll casting process.

Fig. 22- Non-equilibrium fraction solid (Scheil cooling condition) for AZ31 magnesium alloy, graph reproduced
from Reference 19.

Fig. 23- Comparison of predicted and measured temperature for the conditions shown in Table 1, using an HTC
value of 11 kW m-2 °C-1. The solid line shows the perfect match and two dotted lines show ±10% variance form the
correct values.

Fig. 24- Predicted (solid line) and measured (symbol) SDAS (λ) through the strip thickness for a) trial #3 all data, b)
trial #3 averaged data, c) trial #4 all data and d) trial #4 averaged data. The dotted lines show ±20% variance from
the model predictions.

Fig. 25- As-cast microstructure of twin roll cast AZ31 through thickness for trial #3.

Fig. 26- Optical image of microstructure of the strip above the centre-line, the black arrows show the columnar
dendrites growth direction, the circle on lower legends shows where the sample has been chosen.

Fig. 27- Model-predicted effect of casting speed and final thickness on the surface temperature of the exit strip using
an HTC=11 kW m-2 °C-1.

Fig. 28- Model-predicted effect of casting speed and heat transfer coefficient on the surface temperature of the exit
strip for an exit strip thickness of 6 mm.

Fig. 29- Model-predicted temperature gradient at the exit point of the caster for different casting speeds, HTCs and
exit strip thicknesses.

Fig. 30- Model-predicted sump depth for a final strip thickness of 4 mm.

33
 Fig. 31- Model-predicted effect of final thickness on the sump depth for HTC=11 kW m-2 °C-1.

Fig. 32- Model-predicted mushy zone thickness for HTC=11 kW m-2 °C-1.

Fig. 33- Effect of casting speed on the model-predicted cooling rate during solidification for the AZ31 twin roll cast
strip at the surface and centre-line. (Final thickness=4 mm and HTC=11 kW m-2 °C-1).

Fig. 34- Model-predicted effect of casting speed on secondary dendrite arm spacing through the normalised
thickness (0 = top and 1 = bottom) for a final thickness of 4 mm and HTC=11 kW m -2 °C-1.

Fig. 35- Model-predicted effect of casting speed and final thickness on the microstructure uniformity through
thickness for AZ31 twin roll cast strip casting with HTC=11 kW m-2 °C-1.

Fig. 36- Model-predicted effect of casting speed and heat transfer coefficient on the microstructure uniformity
through the thickness of 4 mm AZ31 twin roll cast strip.

34

View publication stats