Continuous casting and mould level control

J. Schuurmans, Corus IJmuiden

A.J.J. van der Weiden, Delft Center for Systems and Control

1
 Continuous casting and mould level
control
This plants discription 1 starts with an introduction to the Corus IJ-
muiden site and the company as a whole. After that, continuous casting is
discussed, which is nowadays used in the production process for steel sheet,
the main product of Corus IJmuiden.
One of the plants where continuous casting is performed is called the
Direct Sheet Plant, which will be discussed in section 3. At the end of this
section the necessity for robust control techniques will be revealed, once
it is shown that the process in the mould of the DSP caster is subject to
time-varying dynamics and disturbances.

1 Corus
Corus is an international metals group that manufactures, processes and
distributes metal products and provides related services such as design,
technology and consultancy. Corus has manufacturing operations in many
countries with major plants located in the UK, the Netherlands, Germany,
France, Norway and Belgium and to provide local service, Corus also has
sales offices and service centers all over the world. Corus is organised into
four divisions and employs 48,500 people in over 40 countries.

1.1 IJmuiden
In IJmuiden, the main site of the Netherlands is located. It is mainly occu-
pied by the strip products division. The manufactured products are high-
grade and coated steels, that exit the plant as rolls.
Approximately 9.500 people are employed in IJmuiden, which produced
6.8 million tonnes crude steel in 2004. IJmuiden.

1.2 Steel sheet production process

The end-product of the IJmuiden site is mainly steel sheet, while coals and
iron ore enter the site. There are several production steps in between, these
will be discussed next.
The raw materials iron ore and coking coal are first treated such that
they are suitable for production in the blast furnace. The coking coal is
1
The plant description and the model is taken from J.P. Kuipers and V.A.H.J. Beentjes
( see references)

2
transformed into to cokes in the coke plant. Iron ore is transformed into
both sinter and pellets in two different plants.
These three components are charged in the blast furnace that melts the
iron ore into iron pig at temperatures between 1000 ◦ C and 1200 ◦ C. This
iron pig is very brittle and is therefore treated in the steel plant. Pure
oxygen is blown over the steel bath and scrap is added; the liquid steel is
then deoxidized and alloyed until specifications are met.
From this point on, the liquid steel can continue on two routes. It can go
to a continuous slab caster, which casts at the thickness of 22.5 centimeters.
The produced slabs are then transported to the hot rolling mill, where the
slabs are reheated and rolled to a pre-determined thickness between 1.5 and
25 millimeters.
Following the other route, the liquid steels goes to the Direct Sheet Plant
which incorporates the casting and rolling process into one operation. The
thickness of the sheet becomes approximately 1 millimeter.
In the cold rolling mills the hot rolled strip is transformed into a finished
product: cold rolled sheet with a thickness of approximately 0.5 millimeter.
Afterwards, extra treatments can take place, such as annealing, coloring and
applying zinc, tin, or chrome coating.

2 Continuous casting
In the beginning of the 20th century, the liquid steel was casted into blocks
(a rectangular box). After solidification of the steel, the steel was lifted out
of the block and reheated such that it could be rolled to obtain steel sheet.
After the 1960s continuous casting became more common; although it
has higher capital costs, the operating costs are lower. It is the most cost
and energy efficient method to mass-produce semi-finished metal products
with consistent quality in a variety of sizes and shapes.
The essence of continuous casting is that the solidification of the liquid
steel takes place on a continuous basis. The liquid steel solidifies against
the mould walls while it is simultaneously withdrawn from the bottom of
the mold at a constant rate. This enables an efficient way to solidify large
volumes of steel. The process works best when all of its aspects operate in
this steady-state manner.
Many different types of continuous casting processes exist. Vertical ma-
chines are used to cast aluminum and a few other metals for special applica-
tions. Curved machines are used for the majority of steel casting. Finally,
thin strip casting is relatively new and has the advantage that the output of

3
the casting can be directly rolled to a thickness of approximate 1 millimeter.

3 Direct Sheet plant

In this section the process that takes place in the Direct Sheet Plant (DSP)
is discussed. Next, remarks are made with respect to quality, output and
process control. To limit the scope of this report, these remarks are restricted
to the influence factors that come from the mould only.

3.1 DSP process plant

The DSP is a relatively young plant. Its building was finished in 2000. The
DSP produces steel sheet out of liquid steel, as explained in section 1. For
this, it uses the thin strip casting technique. The strand, that comes out of
the caster, has a thickness of 70mm. This strand can be rolled out at once,
without the inefficient transportation and reheating that is necessary for the
other production route consisting of the continuous slab caster and the hot
rolling mill.
In figure 3.1, the process of the Direct Sheet Plant is explained in more
detail.

Figure 3.1. Schematic overview of the DSP process.

4
 The liquid steel comes from the steel plant in a ladle [1]. From this ladle
it is poured in a tundish [2], which is essentially a buffer. The tundish allows
for replacing empty ladles without stopping the casting process.
In case the ladle is almost empty, the tundish makes it possible that a
second ladle can take over without having to stop casting.
From the tundish the liquid steel flows through a nozzle into the mould.
This nozzle is submerged in the mould bath. The flow rate trough the nozzle
into the mould can be controlled by a stopper [3] in the tundish. The higher
the position of the stopper, the larger the area through which the steel flows,
the higher the flow rate and vice versa.
The mould [4] is a rectangular copper box without a top and a bottom.
Heat is extracted from the liquid steel through the copper walls of the box
which are water cooled. This causes the outside of the liquid steel to solidify;
an outer shell originates. The mould vibrates vertically to reduce friction
between the solidified outer shell and the copper walls. Also, from above
casting powder is poured onto the liquid steel bath; it melts due to high
temperatures and flows in the gap between the mould and the solidified
outer shell. It forms a lubricating layer which increases the heat flux out of
the liquid steel.
The outer shell is being ’grabbed’ by a driven roll [5] just beneath the
mould, pulling a strand of steel out of the mould. The angular velocity of
the driven roll determines the casting speed. The core of the strand, as it
exits, is still liquid; because of that, the strand proceeds trough a secondary
cooling section [6]. The strand is bend from the vertical plane to horizontal
plane using rolls [7]. At the end of the cooling section the cross-section of the
strand is completely solidified, slabs are then created by cutting the strand
[8].
These slabs are put in a tunnel furnace to let them homogenize [9]. After
a while the slabs come out of the furnace; subsequently they are rolled out,
further cooled and finally coiled [10]; coiled steel sheet is the end-product of
the DSP.

3.2 Quality
After the coiler, the end product of the DSP is tested for its quality which
is done by unrolling a roll and inspecting its surface for pollution or cracks.
Quality problems can be related to the rolling properties, such as temper-
atures and thickness, but can also be traced back to the casting section or
the chemical composition.
As mentioned before, casting powder is poured on top of the steel bath.

5
Ideally, the molten powder (slag) pours down the walls of the mould. How-
ever, a part of the powder solidifies against the cooled walls, above the steel
bath. Occasionally, pieces break off and fall into the steel bath and pollute
the steel.
During casting the level of the steel bath, i.e. the mould level, fluctuates
inevitably. With these fluctuations, the temperatures in the copper walls
near the top of the steel bath vary extremely, which causes the mould to
crack. A crack in the mould is bound to form cracks in the outer shell of
the strand.
In the end, the approval or disapproval of the end product of the DSP
depends on the requirements set. In the beginning of the DSP the require-
ments were mild, but lately, the raised standards have increased the prices
level.

3.3 Output
The output of the DSP plant is of great importance from a business point
of view. Simply reasoned, the more end products that leave the DSP, the
higher its earnings and efficiency.
The output is affected by mould fluctuations for two reasons. First,
if small fluctuations occur, the casting speed is reduced by the operators.
Secondly, high fluctuations can make the operators decide to stop the casting
which leads to longer non-production times and negatively influences the
output.
At the moment, the output is the subject of a discussion within Corus.
A project under the name High Speed Casting is initiated to explore the
possibilities of a higher casting speed, increasing output significantly. A
higher casting speed will definitely worsen the mould level fluctuation, which
will need much attention within the research project.

3.4 Process control

The quality and output issues mentioned above, clearly show the necessity
for good process control. In general, it can be stated that a steady process
and a constant casting speed are desired to ensure a calm initial solidification
of the steel.
Variations in the process disturb the initial solidification and because
of that the strength of the solidified outer shell will not be equal over the
perimeter. If the outer shell contains a weaker spot, the enormous pressure
of the molten steel inside can cause the shell to tear open. This can happen

6
either inside the mould, or even worse, just under the mould causing a
breakout. In that case the liquid steel pours out of the strand, resulting in
damage of the machine and a lengthy standstill.

4 Mould level control

For the quality, the output and the process control it is advantageous to
keep the mould level as constant as possible, this is done by mould level
control. In this section, the configuration of the mould level controller will
be discussed. Afterwards, the current control strategies are discussed.
The parts of the caster that are under consideration for mould level
control consists of a level sensor, the stopper, the nozzle and the mould
itself. Measurements of the mould level are feeded into a control computer,
on the basis whereof control signals are sent out to the stopper. This can
all be seen in figure 4.1.

Figure 4.1. Mould level control

The control computer sends a signal to the actuator. It pushes a lever,

which lifts and lowers the stopper rod. This determines the ingoing flow
rate into the mould through the nozzle. At the bottom a strand of partly
solidified steel is pulled out of the mould by driven rolls. This is seen as the
outgoing flow rate. The outgoing flow rate is almost completely decided by
the operator, who sets the casting speed, which is the angular speed of the
driven rolls.

7
 The net effect of the ingoing flow rate and outgoing flow determines the
level of liquid steel in the mould. The mould level is measured using a level
sensor. For the DSP a Berthold sensor is used. A Berthold sensor consists
of a collector that counts particles that have been sent out by a radioactive
source. Particles are not counted if they are sent into the liquid steel, this
gives a measure which can be related to the level of liquid steel in the mould.
The signal from the sensor is feeded into the control computer. Based
on control algorithms, a control signal is send to the actuator. During a cast
different mould level controllers are used.
In the beginning these controllers were just classical PID controllers.
Additionally, flow controller is used. This control loop uses a sensor that
weighs the tundish and acts on the position of the stopper. If the level of
liquid steel in the tundish increases, the weight will do so as well and this
increase is measured. The higher the level in the tundish, the higher the
ferrostatic pressure at the bottom and the higher the flow rate through the
nozzle (for equal stopper positions). This effect is compensated for by the
flow controller.

5 Robust control
The current controllers are fixed controllers, i.e. they are designed and
implemented in the control computer as a fixed algorithm. However, parts
of the process dynamics and disturbances change over time. If, for instance,
a gain increases in the process, the fixed controllers are not designed for the
higher process gain. This will lead to performance losses or, occasionally,
stability issues.
The parts of the process dynamics and disturbances, that are believed
to be time-varying are listed here:

Stopper During casting, the stopper tip dimension changes. On the one
hand, liquid steel solidifies on the stopper tip and in the nozzle. On the
other hand, the extreme hot flow of liquid steel wears the stopper down.
The extent to which the stopper tip dimension changes depends on
the temperature of the steel and its composition. Besides, aluminium-
oxide, present in the liquid steel, solidifies on the stopper tip, this is
called clogging.

mould-width Also inside the mould liquid steel solidifies and can form a
solid shell. So, the mould-width can change between casts.

8
Standing waves Other research has proven that, especially during start-
up, standing waves in the mould perturb the mould level. These stand-
ing waves are assumed to be a sloshing effect in an open surface, but
the exact nature of the standing waves are unknown.

Bulging Bulging is a the main disturbance in the outflow from the mould,
it takes place in the secondary cooling section . The outer shell is still
thin and the liquid still inside has an enormous pressure. If the strand
has passed a roll, the strand expands, but the strand squeezes again
when in has to pass the next roll. This causes a pumping effect in the
liquid core influencing the outflow rate.

Bad rolls Problems in the secondary cooling section can also cause mould
level disturbances. If one of the rolls is broken or for some reason not
eccentric, it will start to produce vibrations, which propagate into the
mould level.

The nature of all of the above mentioned disturbances is time-varying.

This means that the extent to which they occur depends between different
casting sessions but, moreover, also during a casting session. No models are
present that can predict that extent. Standing waves are currently being
researched. The time-varying nature motivates the choice for resarch into
other control methods.
The goal for this exercise is to design a H-infinity controller, such that
performance improves, with regard to time-varying dynamics and distur-
bances. Not all time-varying elements can be taken into account. A choice
has been made for standing waves, stopper gain and mould-width variations.
The standing waves dynamics are usually mainly seen at the mould level
during start-up. Whether the dynamics have a large influence on the mould
level fluctuations during the bulk is not clear.
For the other two, bulging and rolls, other solutions are also possible.
This issues are still under research.

6 Model of casting process

It is important to decide which parts of the process are incorporated in the
model and which are not. Therefore a system boundary is defined as shown
in figure 6.1. The figure is a modelling representation deduced from figure
4.1.

9
 qi (t) Stopper
System boundary

u(t)
Am

y(t)
Mould

qo (t)
Figure 6.1. System boundary.

The manipulated variable is the stopper position, denoted by u. The

stopper position determines the flow rate through the nozzle into the mould,
qi , which can not measured and is considered an internal variable. The con-
trolled variable y is the measured mould level which is measured by the
Berthold sensor. The flow rate of the strand that exits the mould is consid-
ered the outflow and is denoted by qo . The outflow is mainly determined by
the casting speed vc , which is set by the operators. The cross-section of the
mould is Am ,
From figure 6.1 a basic block diagram for the system can be derived.
This can be seen in figure 6.2. It is a first conceptual block diagram, which
does not include all phenomena yet. This block diagram will be extended.

qo

u qi −
qn ym y
Stopper Mould Sensor
+

Figure 6.2. Basic block diagram for mould process

The block diagram reveals two more internal variables . These are the
net flow rate qn and ym which is the actual mould level that has to be
measured by the sensor.

10
7 Modelling of general process dynamics
In this section, the building blocks of figure 6.2 are being described by
transfer functions.

7.0.1 Mould
The mould can be regarded as a tank. The net inflow qn is the volume
entering per unit time which should be equal to the change per unit time of
the volume Am ym of the mould.
d
Am ym (t) = qn (t)
dt
1
⇔ ym = qn
Am s
where s is the Laplace operator. The resulting transfer function from net
inflow to mould level is:
1
Gm (s) = (1)
Am s

7.0.2 Sensor
The mould level is measured from the bottom of the mould by a Berthold
sensor. It consists of a radioactive source and collector. Due to the use of
radioactivity, the Berthold sensor is known to be influenced by noise n(t).
The noise will be neglected for this excise. With respect to the block diagram
in figure 6.2 the noise signal n(t) acts before the sensor. The transfer function
from actual mould level ym and white noise n to measured mould level y is
given by
1
Gb (s) = (2)
τb s + 1

7.0.3 Stopper and inflow

The stopper gain varies in time, as explained in section 5, whereas the
stopper dynamics does not. The model of the stopper block in figure 6.2
is therefore divided in a transfer function for the stopper dynamics and a
time-varying gain.
The transfer function from reference stopper u position to the actual
stopper position ua is reflected by
1
Gs (s) = (3)
(τs s + 1)2

11
where τs is the time constant. The actual stopper position is subsequently
multiplied by the time-varying stopper gain ks (t) resulting in the ingoing
flow qi
qi (t) = ks (t)ua (t) (4)

7.0.4 Outflow
The outgoing flow, qo , out of the mould is mainly determined by the casting
speed, vc . The casting speeds is set by the operators; typical values are be-
tween 3 and 5m/min. The outflow due to the casting speed, qc , is calculated
as follows:
1
qc (t) = Am vc = 0.1125 vc (t) = 1.875 · 10−3 vc (t) m3 /s (5)
60
The outflow is not just determined by the casting speed, but is modelled by
adding a disturbance flow qd (t) to qc (t)

qo (t) = qc (t) + qd (t) (6)

This disturbance flow will be neglected for this exercise.

7.0.5 Standing waves

Standing waves are the result of interference of surface waves. For example,
they also appear if water is poured into a bucket. Standing waves occurs
mostly during startup of the casting process, but it is believed that standing
waves can also occur during the bulk of casting. Standing waves dynamics
can be represented in the frequency domain by badly damped eigenmodes
with eigenfrequencies at the standing waves frequencies.
The standing waves can be considered a dynamical system on top of
the basic mould process and have eigenfrequencies that are dependent of
the mould width. In order to maintain stability and good performance, the
controller should not excite those eigenfrequencies.
Identification experiments have led to a model for the standing waves.

2a1 b1 ω1 s 2a2 b2 ω2 s
Gw (s) = k1 + k2 2 (7)
s2 + 2b1 ω1 s + ω12 s + 2b2 ω2 s + ω22
The intensities of the standing waves can be altered by the wave gains
k1 and k2 . Reasonable values for the wave gains k1 and k2 are between 0
and 2.

12
 The derived model for the standing waves can be included in the total
model of the mould process. Because of that the block diagram in figure 6.2
is extended with standing waves, this can be seen in figure 7.1.
qo n

u qi −
qn ym +
y
Stopper ks Mould Sensor
+ +

Waves

Figure 7.1. Model of mould process with standing waves.

where also the noise n acting before the sensor is shown.

8 Derivation of total nominal model

In section 7 all the parts of the mould process were modelled separately.
All these parts are combined into one nominal model, with nominal stopper
gain and mould width will be included in the model. For the gains of the
standings waves also nominal values are taken.
It can be seen that the block diagram in figure 7.1 has one output: y.
The stopper position u is the only input signal. The caster speed vc is
known and the other two signals that enter the block diagram, qd and n, are
considered as unknown disturbances.
The model is eight order: the sensor dynamics is first order, the mould is
a simple integrator and the stopper dynamics is second order. The standing
waves are of second order. The model can be written into state space form
as follows:
d
x(t) = Ax(t) + Bu(t) (8)
dt
y(t) = Cx(t) (9)
The matrices A,B,C are given in the matlab file olmodel.mat.

9 Derivation of uncertainty model

The known uncertainties can be used to test robustness of the controller.
The varying stopper gain (50 percent from experiments) and changing mould

13
width (1000-1500mm) can be modeled as multiplicative uncertainty. Which
can be written in the form:

Pp (s) = (1 + wU (s)∆U (s))P (s); |∆U (jω)| ≤ 1∀ω (10)

in which Pp (s) is the perturbed plant P (s) the nominal plant, wU a weigh-
ing function and ∆U any stable transfer function which at each frequency is
less than or equal than one in magnitude (or in block diagram see figure 8.1).
Pp
- wU - ∆U

- P c
-? -

Figure 8.1. Block diagram of transfer function P with multiplicative uncer-

tainty

For finding the weighing function wU we look at the smallest radius lU

which is a maximum to all the possible loop transfer functions Pp :

Pp (jω) − P (jω)
lU (ω) = max (11)
Pp ∈Π P (jω)

Then we choose the rational weight wU which satisfies:

|wU (jω)| ≥ lU (ω) ∀ω (12)

Figure 8.2 shows the calculated radii of the perturbed system for 0.5 <
KS < 1.5 and 1000 < M W < 1500 and the weight function wI .

14
 Bodeplot for finding weight w
I

100

50

−50
Magnitude (dB)
−100

−150

−200

−250

−300

−350

−400
720

540
Phase (deg)

360

180

−180
−4 −3 −2 −1 0 1 2
10 10 10 10 10 10 10
Frequency (rad/sec)

Figure 8.2. Radii of the perturbed system for 0.5 < KS < 1.5 and
1000 < M W < 1500 weighing function wU (thick black line).

The weighing function wU has to be used in the generalized plant model.

References
[1] Kuipers, J.P., Adaptive Mould Level Control for the DSP Caster, Msc
report, 8-4-05.

[2] Beentjes, V.A.H.J., Reducing Mould-Level Fluctuations during Liquid

Core Reduction, Msc report, 24-9-04.

15