Title of Invention

PROCESS AND DEVICE FOR ONLINE ESTIMATION OF NORMAL SHELL THICKNESS AND DEFORMATION IN BREAKOUT DETECTION SYSTEMS

Abstract The invention provides a process for online estimation of normal shell thickness and deformation in a breakout detection system for controlling operations of a steel continuous caster. The process comprises the steps of inputting thermocouple data and online casting parameters to a plurality of fuzzy fault detection modules, and to an intelligent agent for a comparison of current and virtual standard temperatures to identify the instant of updating current standard temperature by the virtual standard temperature. Collating output from said plurality of fuzzy fault detection modules in terms of a value of a breakoutability in a breakoutability analyzer, and evaluating the fault in terms of type and severity of the fault, its location in terms of layer and thermocouple and its evolution with time for presenting to the operator. The invention also provides a device for carrying out the process.
Full Text FIELD OF APPLICATION :
The present invention relates to a process and device for online
estimation" of normal shell thickness and evaluation of deformity
in breakout detection system for controlling operations of a
steel continuous caster. The invention provides a process and
device for predicting shell deformation and its severity of any
type, as they occur, within ambit of a new breakout detection
system in a steel continuous casting process.
BACKGROUND OF THE INVENTION:
The continuous casting process is a fairly recent development in
the history of steel making. Before the advent of continuous
casting, molten steel used to be solidified into ingots in
stationary molds for further processing downstream. The process
renders the need for formation of piecewise ingots in isolated
events as superfluous. Instead, molten steel now continuously
flows into a caster — a copper mold of rectangular cross—section
and about a meter length with cooling water flowing around its
walls and emerges as a solidified slab to be further cut into
discrete lengths.

A major technological bottleneck with continuous casting is the
occurrence of breakouts. When the molten steel cools in the
caster, a semi—solid shell forms at the boundary perimeter of the
moving metal (the strand). This hardens and thickens into a shell
strong enough to support the Ferro—static pressure of molten
steel in the interior, as it emerges out of the caster.
Sometimes, the formative shell tears due to friction with the
mold - leading to leakage of molten steel and disruption of the
casting process. In other cases, the shell develops cracks, OP
remains structurally weak — so that molten steel emerges from the
fault zone and disrupts casting. In either case, this breakdown
of the continuous casting process is called a "breakout" - a very
undesirable phenomena that results in many hours of production
loss before casting is put back on stream.
To prevent occurrence of breakouts in a continuous casting
process, breakout detection and prevention systems were
developed. A few such systems exist globally. These systems rely
on mold wall temperature measurements obtained through
thermocouples embedded in the mold wall in horizontal layers, as
well as on casting variables like speed, composition of steel,
etc. They use various techniques in a processing engine (sitting
on a computing platform) to convert these values, in real time,

into a binary output which says whether a breakout is likely, or
is unlikely, at that instant. This output is passed on to a speed
control device, which reduces casting speed to near zero whenever
a breakout is predicted - in the process healing the deformity.
All occurrence of breakout, as discussed above, can be broadly
classified into two types - those caused by shell tear, and those
resulting from shell weakness like cracks, depressions, thin-
shell deformation, etc. These two types account for more than
95% of all potential breakout, the remaining are due to
operational reasons like start—up, slag entrapments,etc.
Among these two major types of breakout, those caused by shell
weakness - (generally referred to as cracks but also
incorporating depressions, thin-sheels and deep oscillation
marks), account for about 15%.
Most breakout detection systems (BDS) are limited to predicting
and preventing stickers, i.e. breakouts resulting from shell
tear. The reason is that these are very easy to identify from
temperature time—history patterns obtained from the
thermocouples. Usually, there are two horizontal layers of
thermocouples embedded in the mold walls, and as a tear passes

down the mold wall, molten fluid in contact with the wall
produces very high temperature readings on the nearest
thermocouples. This value returns to normal after the tear has
passed. Figure 3 shows four plots, out of these the upper three-
show temperature variation with time for three neighboring
thermocouples in each layer. The higher temperature (usually)
corresponds to a thermocouple from the first layer. The lowest
plot shows casting parameters that are not of interest here. As
the tear passes the two layers in sequence, the 'rising' phase of
temperature at the second layer corresponds to the 'failing'
phase from the first layer, and the temperature—history curves
intersect and cross. This triggers a breakout alarm.
Breakouts resulting from shell weakness are not amenable to such
neat identification. As a consequence no existing BBS worldwide
can identify these with a high level of certainty. Occasionally,
as in the case of a transverse (i.e. horizontal) crack, the
temperature recorded by nearest thermocouple show a dip — due to
an insulation pocket reducing the heat transfer from strand to
mold. This effectively gives rise to an inverse sticker kind of
identification — two upper plots. However, the bulk of cracks are
longitudinal and not transverse, and other cases of shell
weakness, like thin—shells, show no temperature dips at all.

Shell—weakness related breakouts can be further classified into
two - those that result from localized shell deformities, like
cracks, depressions, etc., and those that are characterized by
weakness of the formative shell all round the boundary perimeter.
The latter is usually a consequence of superheat, low mold level,
cooling imbalance or entry of excess slag leading to formation of
a slag layer in the strand. The two types may be termed as lcoal
shell deformities, and 'perimetric* shell weakness.
Perimetric shell weakness is the most difficult to identify, as,
at a given instant, there is no standard to compare with.
Suppose, at a selected horizontal level, the formative boundary
shell is perfectly smooth free of local deformations. What is
the guarantee that for the caster under consideration, at this
casting speed, carbon percentage and horizontal level, the shell
is not thinner than the minimum required for supporting pressure
of molten core as the strand proceeds downwards? In other words,
for a given caster and associated casting conditions, there
exists at every horizontal level a certain characteristic healthy
or normal thickness. If this 'virtual' - as it may not exist at
this moment — normal thickness can be evaluated; it can be used
as a yardstick to measure the current thickness level to evaluate
its instantaneous health. There is yet another issue. Even if

the virtual normal thickness is known, by what mechanism does one
measure the current thickness of such a plastic formative shell
in real time ?
SUMMARY OF THE INVENTION:
The present invention relates to a new and unique method of
predicting shell—deformation and its severity, of any type and as
they occur, within the ambit of a new breakout detection system.
This partially enables the new BDS to progress from the
generation that predicts "stickers", to that which predicts "all
breakouts".
Molten steel proceeding downwards through the cooling mold
gradually forms a thin shell at the boundary perimeter that is in
near contact with mold walls. This shell thickens and hardens as
it moves dawn. Occasionally the formative shell developes
deformities that either intensify or diffuse as the shell
proceeds. In case of intensification, a breakout can occur when a
certain threshold level is crossed.
Deformities can be of various types. An increased insulation
packet between strand shell and mold wall — and moving with the
shell — results in reduced heat transfer in that region leading

possibly to a crack. The crack tends to further increase
insulation between the shell in its immediate vicinity and the
mold wall, leading to its intensification. Alternately, a more
spread-out but less intense insulation zone will lead to weaker
shell formation initially, which can lead to a thin shell or
depression.
There is no known mechanism to measure shell thickness
distribution at any given vertical location (horizontal level)
within a mold, in real time. However, the temperature
distribution at that horizontal level bears a close resemblance
to the shell thickness distribution.
When a crack forms, longitudinal or transverse, the mold wall
temperature will be lower and a thermocouple embedded at that
location will exhibit the same. A similar condition will be
observed for a depression. In case of a thin shell, the molten
steel within the shell is closer to the mold wall, as a
consequence the corresponding thermocouple will show a higher
temperature.
Reference to "lower" or "higher* temperatures imply there is a
'standard* temperature against which the comparison is being

made. This standard temperature is that which corresponds to the
'average* shell thickness, i.e. if the existing deformities at a
horizontal level were, for a moment, all straightened out — then
the thickness that would result. It then implies that the
standard temperature, analogously, may be obtained simply by
taking the mean of temperatures across all thermocouples in a
layer. Just as the shell thickness deviates from the "standard
thickness* wherever deformities are present, the local
thermocouple temperature will also deviate from the 'standard
temperature* wherever deformities are present, further, this
mapping is direct, i.e. greater the local deformity, higher the
temperature deviation. Also, this deviation is descriptive, i.e.
negative deviation implies cracks, and positive deviation denotes
thin—shells.
As the strand moves down in the mold, the 'average' temperature
falls, i.e. the mold wall temperature at lower horizontal levels
are lower. This is because the temperature gradients between
strand shell and the flowing coolant decrease as the shell
transverses the mold length and cools. Correspondingly, the
reflection of shell deformities into local temperatures also get
damped in magnitude (in either positive or negative direction)

at lower horizontal layers. A simple way to neutralize this
factor - the level effect — is to consider the ratio of local
thermocouple temperature to average (i.e. horizontal level)
temperature, rather than the directed difference.
In the present invention the ratio of local thermocouple
temperature to average layer temperature is used as an indicator
of shell deformity in the region corresponding to that
thermocouple.
The above aspect of the invention provides a measure of the
deviation from 'normal' , it does not provide the 'normal' shell
thickness.
A statistical model developed by Vaculik et al (US Patent
6564119) provides 'normal casting conditions' for a known caster
and casting parameters; the measure of 'normal shell thickness'
can be considered as an element of the total set of 'normal
casting conditions". However, to extract the "normal casting
conditions' for a given casting mold, at a given casting speed,
chemical composition, lubricating powder condition, etc., it
takes a few months of data generation from online casting. This
has two disadvantages, first, it cannot be implemented from the
word 'go' in a new caster, and two, whenever any of the

parameters of the model (say speed) occupy values outside its
range specified in the data generation phase, the results are
likely to be misleading. This brings us to the other aspect
of the invention.
The other aspect of this invention relates to the establishment
of a technique to acquire the "normal shell thickness' with
immediate effect, irrespective of caster characteristics and
casting parameters. This enables — in company with the first
aspect of the invention — the new breakout detection system to
measure the degree of deviation of shell perimetric thickness
from normal, in real time, and predict the intensity of fault due
to shell structural weakness, without any overheads like prior
data generation and parameter range limitations.
The present invention provides a process for online estimation of
normal shell thickness and shell deformation in a breakout
detection system for controlling operations of a steel continuous
caster, said process characterizead by inputing thermocouple

data and online casting parameters to a plurality of fuzzy
fault detection modules and to an intelligent agent for a
comparison of current and virtual standard temperatures to
identify the instant of updating current standard temperature by
the virtual standard temperature; collating output from said
plurality of fuzzy fault detection modules in terms of a value of

a breakoutability in a breakoutability analyzer; and evaluating

the fault in terms of type and everity of the fault, its location in

terms of layer and thermocouple and its evolution with time for presenting
to the operator.
The present invention also provides a device for online
estimation of normal shell thickness and shell deformation in a
breakout detection system for controlling operations of a steel
continuous caster, said device comprising thermocouples arranged
in mold walls; a plurality of fuzzy fault modules arranged in
parallel and an intelligent agent, said plurality of fuzzy fault
detection modules and said intelligent agent for receiving data
from thermocouple and online casting parameters; and a
breakoutability analyzer for collating the outputs from said
plurality of fuzzy fault detection modules in terms of a value of
breakoutability and for evaluating the fault in terms of severity
of the fault, its location in layer or thermocouple and its
evolution with time for presenting to the operator.
BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWIN6S
In the accompanying drawings :
Figure 1 shows configuration of a slab caster machine;
Figure 2 shows the schematic of thermocouple layout in a typical
mold with two layers;

Figure 3 shows the temperature time-history pattern of a sticker;
Figure 4 shows the temperature time—history for a crack;
Figure 5 shows the breakout detection system architecture;
Figure 6 shows the fuzzy membership function distribution for
temperature ratios;
Figure 7 shows the fuzzy membership function distribution
identical for both normalized inputs CST and VST;
Figure 9a, 8b and 8c show a flow chart of the high level breakout
detection system.
In case of a true deformity in the shell, the temperature of the
corresponding thermocouple will be significantly different from
the layer average, and will tend to adversely influence the
average and damp its own severity. Hence, when checking each
thermocouple for a deformity at its corresponding shell location,
the average is calculated without consideration of that
thermocouple, i.e. the average is really 'average minus
thermocouple of interest'.
A wider picture of the present invention can be provided in terms
of its role within a continuous caster monitoring and breakout
detection system. Figure 5 shows the conceptual architecture of
this system. Thermocouple inputs 1 and the online casting
parameters are fed to three fuzzy fault detection modules 2,3,4

arranged in parallel. Thermocouple inputs are also provided to
an intelligent agent 5. The three fuzzy fault detection modules
are sticker detection module 2, crack detection module 3 and thin
shell detection module 4. Out of these modules the crack
detection module 3 evaluates crack and local thin shells. The
present invention fits in this crack detection module 3.
Outputs from all these fuzzy fault detection modules 2, 3, 4 in
terms of value of breakoutability that effectively represents the
supplement of degree of normality in a scale of 0 to 100 is
collated in a breakoutability analyzer 6. The breakoutability
analyzer 6 evaluates the type and severity of maximum fault,
representing a sticker, a crack, local or perimetric thin shell,
etc., its location in terms of layer and thermocouple and its
evolution with time and presents it to the operator. It may be
noted that the input—output feed—forward cycle operates in real-
time .
Fuzzy logic is considered for all the fault detection modules as
this provides a natural way to convert causes of fault into
intensity of fault, instead of the usual breakout/no—breakout
output paradigm followed in nearly all extant breakout detection

systems. In the module of interest, i.e. crack detection, the
fuzzy system takes four inputs. Processing control flows over
each layer and across each thermocouple in a layer, and the fuzzy
analysis module is executed therein for each thermocouple. The
output, i.e. breakoutability, again from each thermocouple, is
sent to the breakoutability analyzer.
The first and most significant of the four inputs to the fuzzy
analysis module is the ratio of local thermocouple temperature to
the instantaneous average temperature. The other inputs are of a
supportive nature and help in refining the value of
breakoutability. These are, first, the casting speed, second, the
percentage of Carbon, and third, the vertical location. All the
fuzzy inputs have a range of existence that depends on physical
conditions. These are mapped into the domain {-1:1} far
fuzzification using a quadratic transformation. Table 1 describes
the four variables, their range of existence and normalization
mechanism, and the type and number of fuzzy membership functions.



The first input, i.e. temperature ratio, is the driver of
breakoutability and also the core of this invention. The
membership functions of this variable are shown in more detail
in Fig.6.
With reference to the other aspect of this invention, in a
continuous casting process for steel, the solidification of
liquid steel initiates at the strand boundary perimeter - the
region nearest to cooling mold walls — and advances inwards even
as the strand moves down. For the casting process to be stable, a
close relationship needs to be maintained between the mold
coolant flux, the casting speed, and the liquid steel entry
temperature, with the driving objective that the shell thickness
growth rate must be sufficient to be able to support ferro—static
pressure of molten core in strand, near the mold bottom.
Disturbance in the equilibrium between the above three parameters
or the return towards equilibrium from perturbations caused by
variation in other casting parameters, will affect shell
thickness growth - sometimes to the extend that the driving
objective mentioned above may not be attained. This will lead to
a breakout.

The best way to prevent such a breakout would be to pick up the
shell thickness at any horizontal layer of interest, and compare
it with an available "normal thicknes'. This "normal", is of
course, a strong function of many casting parameters — as
discussed in the last section.
In following such an approach, the first concern is that there is
no available means to pick up shell thickness in real time.
However, the average shell thickness around boundary perimeter at
a given horizontal layer bears a close resemblance to the average
temperature of all thermocouples embedded in mold walls at that
horizontal level. If the shell becomes thin, then the molten core
will be closer to mold wall than usual,and the observed average
temperature will be high. The converse for a thicker shell.
Also, if a slag layer entrapped in the strand and passes a layer
of thermocouple, the reduced heat transfer will reflect in a
lower average temperature. Hence, there is a direct relationship
between extend of deviation of shell thickness from 'normal
thickness', and the degree of divergence of average layer
temperature from a 'normal layer temperature'. In other words,
one can transform the analysis of shell thickness deviation from
normal thickness at a layer, into an analysis of layer average

temperature deviation from layer normal temperature. This
transformation, and its implementation in the new caster
monitoring and breakout detection system, constitute the first
aspect of this invention.
The second concern is that the 'normal layer temperature* is as
etheral an entity as 'normal layer thickness'. Branted that
temperatures can be obtained easily from the thermocouple while
layer thickness cannot be obtained at all. However, the 'normal
layer temperature* is as virtual as 'normal layer thickness' - in
the sense that it does not physically exist (unlike instantaneous
average temperature) at that point in time. This etheral entity
is brought into the inundate plane, i.e. made from virtual, using
an intelligent agent — the second part of this invention.
For ease of discussion we refer to the 'normal layer
temperature" as current standard temperature or CST. As
discussed, it is a function of a large number of casting
parameters. Under transitory conditions it follows the parameters
with a lag, ie. the CST at agiven instant is that which pertains
to parametric states at ∆t earlier. When conditions stabilize,
the CST stabilizes after time ∆t.
The intelligent agent assumes, firstly, that for a given set of
values of all casting parameters in a caster, there exists a
unique CST. Secondly, the best measure of CST is an average

across time of the instantaneous averages across thermocouple
(see equation 1 below) of temperature, when casting conditions
are stable. The ingenuity of the intelligent agent lies in
moving from one set of casting conditions to another, through a
transitory phase.
To achieve this transition, a virtual standard temperature or VST
is defined. This VST is obtained using the above average, i.e.

where i denotes summation over past N time steps, j the summation
M number of thermocouples at each time step, and T the
temperature. If the instantaneous layer average changes
suddenly, the VST will reflect this change, but slowly - like the
lag in real. If that average changes slowly, the VST will change
even more slowly.
Thus at any given instant the intelligent agent has a CST and a
VST. The CST is the working value of standard temperature - what
the caster monitoring and breakout detection system is using at
that instant. The VST is a potential CST in the background. Under
stable conditions, they match exactly. In transitory states, VST

starts deviating from CST. The agent has to decide exactly, when
the CST has to be replaced by the VST. That is done by decision
making element of the intelligent agent.
The decision making element of the intelligent agent is a Takaji
Sugeno type fuzzy system. It takes two inputs with a single input
- to replace or not to replace. The first input is the absolute
value of the difference betwen the CST and VST. Obviously, the
larger this value, the larger the compulsion for change. As is
any fuzzy system the absolute difference is normalized to {-1:1}
through a transform, refer to Table 1. The second input is the
absolute value of the rate of change of VST. In fact, it is the
temporal average of absolute values of rate of change of VST,
taken over a few time steps. When the rate of change of VST is
fluctuating, i.e. casting parameters are not changing
monotonously, the replacement of CST is impeded. In other words,
the CST should not chase such a turbulent VST. When this rate is
very low, i.e. VST has stabilized, or is high and monotonous,
i.e. VST is rapidly changing to a different level, the change of
CST is facilitated. Figure 6 shows the fuzzy membership function
distribution of the two inputs — also refer to Table 2.



Using this novel approach of picking up normal shell thickness
direction from the dynamic casting environment, the need to frame
a model based on casting parameters and subsequently train it on
a running caster, is made redundant.
The evaluated values of normal temperature (current standard
temperature) for each layer of thermocouples emanating from the
intelligent agent are used in the thin-shell prediction module,
where these are combined with the instantaneous average
temperatures of the corresponding layers to generate the degree
of deviation of shell thickness from normality (refer discussions
above). This deviation is the most significant input in the
fuzzy thin-shell module, where it is combined with three other
inputs - the casting speed, percentage of Carbon, and layer
vertical position in mold — to generate the breakoutability.
The present invention has enabled the new breakout detection
system to evaluate in real time and with very good accuracy the
nature and intensity of local shell deformities of any kind, as
well as the shell perimetric weakness at horizontal levels of
interest using the simple well—known mechanism of thermocouple
layers at these levels. This facilitates online shell health
monitoring with the provision for raising breakout alarms
whenever a fault crosses a threshold.

WE CLAIM
1. A process for online estimation of normal shell thickness and deformation
in a breakout detection system for controlling operations of a steel
continuous caster, said process characterized by the steps of:
- inputing thermocouple data and online casting parameters to a
plurality of fuzzy fault detection modules, and to an intelligent
agent for a comparison of current and virtual standard
temperatures to identify the instant of updating current standard
temperature by the virtual standard temperature;
- collating output from said plurality of fuzzy fault detection modules
in terms of a value of a breakoutability in a breakoutability
analyzer; and
- evaluating the fault in terms of type and severity of the fault, its
location in terms of layer and thermocouple and its evolution with
time for presenting to the operator.

2. The process as claimed in claim 1, wherein the evaluation of severity of
the fault is in terms of evaluation of sticker, crack, local or perimetric thin
shell.
3. The process as claimed in claim 1, wherein collating the outputs from said
plurality of fuzzy fault detection modules is in terms of a value of
breakoutability that effectively represents the supplement of degree of
normality in a scale of 0 to 100.

4. The process as claimed in claim 1, wherein the fuzzy inputs to said crack
detection module comprises ratio of local thermocouple temperature to
the Instantaneous average temperature, casting speed, percentage of
carbon and vertical location.
5. The process as claimed in claim 4, wherein the ratio of local thermocouple
temperature to the instantaneous average temperature is used as an
indicator of shell deformity in the region corresponding to that
thermocouple.
6. The process as claimed in claim 1, wherein the determination of the
severity of fault is by combining the evaluated value of normal
temperature (CST) for each layer of thermocouples with the instantaneous
thermocouple layer perimetric average temperature.
7. The process as claimed in claim 6, wherein a multi-variable comparison of
some functions of the current and virtual standard temperatures, i.e. CST
and VST, are made in the intelligent agent to decide on the instant of
updating of CST by VST, by using a fuzzy system of Takagi Sugeno type.
8. A device for online estimation of normal shell thickness and shell
deformation in a breakout detection system for controlling operations of a
steel continuous caster, said device comprising:

- thermocouples arranged in mold walls;
- a plurality of fuzzy fault detection modules arranged in parallel and
an intelligent agent, said plurality of fuzzy fault detection modules
and said intelligent agent for receiving data from thermocouple and
online casting parameters; and

- a breakoutability analyzer for collating the outputs from said
plurality of fuzzy fault detection modules in terms of a value of
breakoutability and for evaluating the fault in terms of severity of
the fault, its location in layer or thermocouple and its evolution with
time for presenting to the operator.
9. A process carried out with the help of a device for online estimation of
normal shell thickness and shell deformation in a breakout detection
system for controlling operations of a steel continuous caster, substantially
as herein described and illustrated in the accompanying drawing.

Documents:

420-KOL-2004-ABSTRACT.pdf

420-KOL-2004-CLAIMS.pdf

420-kol-2004-correspondence-1.1.pdf

420-KOL-2004-CORRESPONDENCE.pdf

420-kol-2004-examination report.pdf

420-kol-2004-form 1-1.1.pdf

420-kol-2004-form 1.1.pdf

420-KOL-2004-FORM 1.pdf

420-kol-2004-form 13.pdf

420-kol-2004-form 18.pdf

420-KOL-2004-FORM 2.pdf

420-kol-2004-form 3-1.1.pdf

420-KOL-2004-FORM 3.pdf

420-kol-2004-form 5.pdf

420-kol-2004-gpa.pdf

420-kol-2004-granted-abstract.pdf

420-kol-2004-granted-claims.pdf

420-kol-2004-granted-description (complete).pdf

420-kol-2004-granted-drawings.pdf

420-kol-2004-granted-form 1.pdf

420-kol-2004-granted-form 2.pdf

420-kol-2004-granted-specification.pdf

420-kol-2004-others-1.1.pdf

420-KOL-2004-OTHERS.pdf

420-kol-2004-reply to examination report.pdf


Patent Number 247790
Indian Patent Application Number 420/KOL/2004
PG Journal Number 20/2011
Publication Date 20-May-2011
Grant Date 18-May-2011
Date of Filing 16-Jul-2004
Name of Patentee TATA STEEL LIMITED
Applicant Address JAMSHEDPUR
Inventors:
# Inventor's Name Inventor's Address
1 ARYA, K. BHATTACHARYA TATA STEEL LIMITED, JAMSHEDPUR-831 001
2 K. CHITHRA TATA STEEL LIMITED, JAMSHEDPUR-831 001
3 S.S.V.S. JATLA TATA STEEL LIMITED, JAMSHEDPUR-831 001
4 P. SRINIVAS TATA STEEL LIMITED, JAMSHEDPUR-831 001
5 VINAY MAHASABDE TATA STEEL LIMITED, JAMSHEDPUR-831 001
PCT International Classification Number G01K7/00
PCT International Application Number N/A
PCT International Filing date
PCT Conventions:
# PCT Application Number Date of Convention Priority Country
1 NA