FIELD OF THE INVENTION:
The present invention relates to an on-line method and system for advance
prediction of lumb-type break-outs to initiate preventing measures in a
continuous casting process of steel manufacturing chain using thin-type slab
casters.
BACKGROUND OF THE INVENTION:
The steel making chain of processes initiates at the mineral extraction and
refinement stages, passes through coke and sinter making, then making of liquid
iron in the blast furnace followed by its transformation to liquid steel through
chemical processing. The next major step is the continuous transformation of the
liquid steel to solid, followed by thermo-mechanical processing of discrete solid
slabs (or blooms/billets) into thin sheets of different properties. The step of
continuous transformation of liquid steel to solid is most critical amongst the
entire process-chain, from the perspective of product quality including process
stability and throughput. Any disruption, delay or slackness in this stage
immediately influences the upstream liquid including the downstream solid
processing. The step of continuous transformation is called continuous casting
which is further described in prior art references (1, 2).

One of the typical and most undesirable disruptions in the continuous casting
process is occurrence of "breakouts" in the mould. This phenomenon occurs
when the thin solid shell that develops all around the perimeter of the
continuously cast strand, and which thickens with the progression of cooling, and
suddenly breaks resulting in discharge of the liquid steel onto the downstream
machinery and equipment. Such undesired phenomenon effectively disrupts the
casting process for a few hours. However, breakouts in the mould might entail
one or more causes for example, primarily stickers, cracks and improper shell
formation. Their relative frequency depends on the specific nature of the casting
process.
Conventional slab casters (also called thick slab casters) generally have slab
sectional thickness of around 200-250 mm and cast at speeds between 1.2-2.2
meters per minute. In modern thin slab casters (figure 1 provides a schematic of
a known thin slab casting line), the slab sectional thickness and cast-speed vary
between 75-95 mm and 4-6 m/min respectively. Breakouts in the former type of
slab casters of "sticker" type where formative thin shell sticks to the mould wall
and the surrounding region is ruptured due to tension. In the thin type casters,
breakouts are predominantly caused by "cracks" and other shell formative

defects like "lumb" where the shell is weak at the narrow faces, bulges and
breaks with the increase in ferro-static pressure.
In order to eliminate the disruption in the steel making chain caused by
breakouts, prior art provided methods to detect in advance formation of
breakout conditions and initiate preventive measures. For this purpose, mould
walls embedded with arrays of horizontally layered thermocouples (TC), was
proposed, wherein each TC reflecting the instantaneous variation in temperature
of strand shell closest to the sensing point. It is known that "Stickers" are
accompanied by a typical rise and then fall of temperature as the "tear" passes
the point, and followed by a corresponding rise and fall in the TC disposed
immediately below. This typical pattern with occasional (sometimes severe)
variations is detected by a pattern-recognition software, upon detection of
formation of a tear likely to result in a breakout, an alarm is sounded. The
casting speed is then reduced which allows the fault (tear) to heal and thus
prevent the sticker breakout.
While the TC based breakout detection system works well in conventional
casters, they do not adequately serve the purpose in thin slab casters, mainly for
the following two reasons:

(1) Breakouts of sticker type constitute only a minor proportion of all the
possible types of breakouts in the thin slab casters, where the other types
of breakout are more common, and which are not accompanied with neat
temperature-time patterns as seen in stickers type.
(2) Due to higher casting speeds, the residence time of a fault in the mold,
after its detection by TC, is only around 10 sees. The speed reduction
takes effect in around 6 sees, leaving very little time for actual healing of
the fault even after detection. Thus even if a fault is detected by the TC in
the mould, prevention of the resultant breakout cannot be ensured.
These two factors make conventional TC-based breakout detection method
unsuitable for thin slab casters, and accordingly, a need arises to propose an
alternative detection method.
The second factor as discussed herein above, warrants that the fault be detected
even before it has formed in the mould, because once formed, there will be a
little time left to prevent the break-out. This eliminates the possibility of using a
TC for direct one-to-one mappings between temperature-time patterns and
crystallized faults. Furthermore, according to prior art, the fault detection can
only be made by recognizing some typical emergent pattern prior to fault
formation in a few among hundreds of online parameters that characterize the

continuous casting process. It is assumed that a fault formation is preceded by,
or in other words the root cause of a fault lies in certain specific combinatorial
pattern of parametric data representing a few among the hundreds of
parameters characterizing casting. Further, such a specific combinatorial pattern
needs to be recognized in real time. Over and above, it is necessary that on-line
of a specific combinational pattern be accurate in the sense that genuine faults
are substantially captured, and that the pattern-recognition is enabled without
raising false alarm. And the alarming has to be sufficiently in advance to allow an
ensured operator response for mitigation.
Examples of prior work on thermocouple based breakout detection systems may
be seen in IPA, 420/KOL/04 and 773/KOL/2011. Prior art based on multi-
parametric analysis are disclosed in US patent 6,564,119 and patent publication
US 2004/0172153. In the latter, statistical techniques are used to evaluate the
difference or 'distance' between a set of running parameters of casting and a
corresponding set of ideal parameters extracted from best conditions. The
distance provides real time monitoring and when it exceeds some threshold, an
alarm is raised. In this approach an ideal condition is not taken into account as
that itself is a function of other casting characteristics, including experimentation
with new process conditions. Instead running conditions of casting are taken as

baseline and significant deteriorations are predicted by the ANN are considered
as digressions which may lead to breakout. While the ANN output can be viewed
as a monitoring parameter, severe digressions are alarmed for immediate
responsive action.
As explained herein above, the continuous casting process is one of the most
important links in the steel making chain as it represents a transition state,
where the steel is transformed from liquid to solid state, which transition state
highly influences the stability, productivity & quality. One of these constraints
relates to "breakouts" where the formative solid shell encapsulating the liquid
suddenly ruptures in the mould leading to drainage of the internal liquid and
consequent blockage of equipment and process. In the thick (conventional)
casters, 'stickers', i.e. shell sticking to mould, are as explained earlier is the
predominant cause of breakouts. Accordingly, such casters are provided with
'Breakout Detection Systems' (BDS) that tap signals from an array of mold-
embedded thermocouples and process the signals in real time to recognize
specific patterns that precede sticker breakouts. On recognition, an alarm is
generated with consequent reduction in casting speed and mitigation of fault. In
thin slab casters, conventional BDS is also defunct due to the fact that 'stickers'

constitute only a minor proportion of all the possible breakouts. Further, casting
speeds in thin type slab casters are so high that the time between sensing of a
fault by the thermocouples and its mitigation by speed reduction of the casters
is insufficient for complete cure of the ruptured steel, and results in breakout.
OBJECTS OF THE INVENTION:
It is therefore an object of the invention to propose an on-line method and
system for advance prediction of lumb-type break-outs to initiate preventing
measures in a continuous casting process of steel manufacturing chain using
thin-type slab casters.
Another object of the invention is to propose an on-line system for advance
prediction of lumb-type break-outs to initiate preventing measures in a
continuous casting process of steel manufacturing chain using thin-type slab
casters.
SUMMARY OF THE INVENTION:
According to the invention, specific combinatorial pattern relating to a lumb type
of fault is recognized in process real time is performed by an Artificial Neural

Network (ANN). Further verification of prediction to filter out possibilities of error
is performed by using modeling techniques. According to the invention, an
advance prediction system for faults in thin slabs casters is proposed which is
enabled to detect the faults characteristics to such casters. Further, a sufficiently
advance prediction enables successful mitigation by responsive action. In
particular, the inventive system is applicable to breakouts of xl_umb' type. The
invention allows input of online multiple casting process parameters into an
Artificial Neural Network (ANN) which delivers in real time a value of 'Abnormality
Index' (AI). This AI is further tracked to recognize sudden and continuous
increase indicative of sharp deterioration of the casting and an impeding fault,
followed by generation of an alarm and mitigating action. The system has been
validated for high accuracy, i.e. capturing most faults amidst low false positives.
The operation of the inventive system can be sequentially splitted into two
phases. In a first phase, an offline-synthesized subset of the total set of process
parameters is fed as input into the ANN which generates a dimensionless number
that represents the instantaneous quality of casting. This number varies from 0
to 1 where the former denotes very good casting and the latter a breakout.
Hence the number can be termed as an instantaneous 'Abnormality Index', (AI),
as it represents the degree of abnormality in the casting. This instantaneous AI
can fluctuate within short time spans but remains within range, hence it is more

appropriate to deal with the short-time moving average of this AI, termed MAI.
In the second phase, the MAI is transformed into an alarming index using
modeling techniques utilizing the mid-term historical variation of this MAI in the
current casting sequence. The alarming index (AI) is usually 0 but whenever it
reaches 1, an alarm is raised and preventive action initiated.
BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWINGS:
Fig. 1. Schematically illustrates a complete thin slab casting process of prior art.
Fig. 2. Basic Layer and Neuronal structure of an Artificial Neural Network (ANN)
adapted in the invention.
Fig. 3. Flow of computation in an Artificial Neural Network of Fig. 1
Fig. 4. Flowchart detailing the flow of logic in the advance alarming system
according to the invention.
Fig. 5. High Level Architecture of the Advance Alarming System of the invention.

DESCRIPTION OF A PREFERRED EMBODIMENT OF THE
INVENTION:
This invention provides real time analysis of the combination of online
parameters that accompany continuous casting, particularly thin slab continuous
casting, and consequent decision making in real time as to whether or not, a
fault condition is impending which can be mitigated by taking preventing action.
An Artificial Neural Network, from a mathematical perspective, is a 'Universal
Approximator'. This implies that, given a mathematical function of any degree of
complexity, the ANN can emulate this to any desired degree of accuracy [Ref].
By the feature "emulate", it is implied that, given a set of input values to both
the ANN and the mathematical function, the output from the ANN shall be
sufficiently close to that provided by the closed-form transformation.
Generally, an ANN is used when there is some amount of data available from a
process or system that intrinsically encapsulates the functionality of that process,
and an user captures this functionality. An ANN precisely serves that purpose.
This data is used to 'train' the ANN so that it learns this functionality, and when a
new input vector is provided to the ANN so that it learns this functionality, and
when a new input vendor is provided to the ANN it emulates the process to

generate representative output. The core of the training process is the iterative
readjustment of the intra-neuronal weights so that the final output generated by
the ANN for a given input vector comes as close as possible to the actual process
output for the same input vector, and this holds true for each of the training data
records.
Structurally an ANN is made up of at least three layers of neurons, where the
first is the input layer, the last is the output layer, and the rest is the hidden
layer. This is shown in figs. 2 & 3. Each neuron in a layer is connected, in
principle, to all neurons in its neighboring layer/s. The neuron first sums up all
inputs from its preceding layer neurons multiplied by the connecting weight
between the two neurons, and then provides a nonlinear transform to the sum.
This becomes it output which now serves as input to the neurons of the next
layer. The summing up relationship is

and the nonlinear transform is of type


in the above relations wik represents the weight of the connection between
neurons k in the previous layer and i in the current, and the nonlinear function
shown is the sigmoid function as an example. The outputs {y} from the first or
input layer are simply the process input values, and the outputs from the output
layer are the final process outputs corresponding to the given inputs. As is seen
clearly from fig. 3, the flow of information is from left to right across the layers.
The ANN of this invention is inputted with the process parameters of continuous
casting as the inputs, and generates as output an abstract or 'conceptual'
quantity for example, the Abnormality Index. In a sense it monitors the health of
casting. The Abnormality Index is termed "abstract or conceptual" as it is not a
physical quantity that can be measured. There does not exist any process output
data like "abnormality of casting" on which the ANN can be trained. Instead, the
data available is binary - either 0 implying good casting or a 1 implying
breakout. Hence by the principles of training, the ANN also generates a binary
output - either a 0 or a 1.
However, the problem is that the quality of the casting process is not binary
classifiable, in terms of either perfect or a breakout. Most of the time, it is in
between, and much closer to 0 than to 1. No direct means being available to

measure the actual state of a casting process to train the ANN, except that,
which is available from the process is a 0 or a 1. More importantly, the process
shifts from state 0 to state 1 continuously and almost monotonously in the few
minutes preceding a breakout. Thus, the invention captures this shift in the
process, by using the third element of information on the process, apart from a 0
or a 1. This is the fact that the shift from 0 to 1 is near continuous and near
monotonous. So the output from the ANN, which is also either close to 0 or close
to 1 reflecting its training master (real process) output, is continuously tracked to
identify periods of continuous and monotonous increase, and then decides
whether a situation has been identified synonymous to that prevailing just a few
minutes before a breakout.
In order to achieve the pre-requisites as stated herein above, firstly a moving-
average AI or MAI is used. Secondly, instead of using instantaneous AI. As a
continuous casting process is associated with more than three hundred online
process parameters, the ANN is not expected to handle that sort of huge input
numbers; additionally all calculations need to be in process real time. Hence,
through careful analysis that combines core domain knowledge on the
continuous casting process with statistical and heuristic methods, the number of
inputs actually provided to the ANN is reduced to only six (6) for example;

(1) Absolute value of time derivatives of level summed over previous thirty
time steps
(2) Maximum (of two) wide face heat fluxes
(3) Absolute value of difference between two wide face heat fluxes
(4) Signed difference between wide face average and narrow face
maximum heat fluxes (note that all face heat fluxes are normalized
w.r.t. area)
(5) Average (of two) friction values
(6) Absolute value of time derivatives of averaged (as in #5) friction.
To arrive at a conclusion, whether the current conditions qualify for raising an
alarm, the most important factor is a steep rise in MAI. For identifying such a
steep rise, a first step is to create a baseline value that represents a quality
casting in the current sequence. This is termed as NICC or 'Normality Index of
Current Casting'. It is extracted by taking a long-term average of the value of
MAI. Next, the running MAI is compared against the NICC, and checked if a
threshold value of the ratio of these two quantities has been crossed. If the NICC
is very close to zero (as can happen very often), this threshold ratio is crossed
very easily. To prevent this occurrence, a minimum alarming threshold of MAI is

also defined. If both these conditions are met, it is to be considered that a
potential alarming situation is created.
A potential alarm is converted to an actual alarm after a few more checks. First,
there is a 'reaction period' after an earlier alarm has been raised and mitigating
action has been taken, and the situation is still to stabilize. An alarm is not raised
within this 'shadow window' of a previous alarm. Second, a check for continuity
and monotony is made, i.e. the 'potential alarm' is allowed to continue for at
least ten seconds (the 'threat thresholdO before the actual alarm is raised.
There is an alternate situation where the NICC is running at high levels.
Imposing the 'threshold ratio' on this high NICC implies pushing the requirement
on MAI to be very high, which would lead to missing genuine alarms. This
usually occurs when the sequence has started shortly and the time window for
calculating NICC is small. For eliminating this possibility, an upper threshold,
called 'NICC threshold' is imposed on NICC - if that value is crossed, NICC is set
to that value.
The steps for alarm generation in the system is illustrated through the flowchart
of fig. 4, for example, :

1.0 The ANN runs on a processor that receives the field inputs in real time
and generates an Al every 1 sec
2.0 The value used for analysis is not the instantaneously predicted Al but the
moving average taken over a period of 10 seconds termed as MAI
3.0 At every instant there is associated a Normal Index of Current Casting
(NICC) which can be an average value of Al of past one hour of casting or
(if casting is recently started) a value from lookup tables based on casting
product (explanation follows below)
4.0 The Al output is compared against the NICC and if it is some threshold
ratio (or 'Rise FactorO more than NICC it is taken as digression into bad
casting
5.0 The software on processor then looks at the values of different input
variables and uses a system of logic to decide the choice of the remedial
action
6.0 The remedial action is flashed on HMI screens for operators to execute
Explanations:
for #3: The NICC is a baseline Al value predicted by the ANN over periods of
quality casting. In the 168 hours of the week, it may be assumed that

approximately 160 hours of quality casting is performed. In other words,
95% of the time casting is "quality casting". However, the value of AI
can drift based on specific casting conditions. Hence, the average AI of
last one hour of casting can be considered as the reference baseline
Abnormality Index of quality casting is called Normal Index of Current
Casting or NICC.
for #4: as explained earlier, when casting (as an operation) quality deteriorates
to a level where a LFC corner breakout becomes likely, and AI (moving
window average over 10 secs) starts moving steeply upwards while the
NICC (moving window average over 3600) remains practically
unchanged. Then the ratio of AI to NICC, which would normally be close
to 1, also shoots up sharply. If the ratio crosses the 'Rise Factor', it is
taken indicative of a breakout if an additional criterion is also met. This
is simply that the AI should be above a threshold value called the 'MAI
Threshold'.
The architecture of the Advance Alarming System incorporating the programme
module is illustrated in fig. 5.


Other References
1. World Steel University website, continuous casting link:
http://www.steeluniversity.org/content/html/ena/default.asp?catid=27&paaei
d=2081271519f valid Mar 23, 2011.
2. B.G. Thomas, "Modelling of the Continuous Casting of Steel-Past, Present
and Future", Metallurgical and Materials Transactions B, Vol 33B, No. 6,
Dec 2002, pp. 795-812.
3. Kurt Hornik: Approximation Capabilities of Multilayer Feedforward
Networks, vol. 4,1991.

WE CLAIM:
1. An online method for advance-prediction of "lumb" type break-outs to
initiate preventive measures in a continuous casting process of steel
manufacturing using thin-type slab caster, the process comprising the
steps of passing the molten steel through a water-cooled substantially
vertically-aligned lubricated mould to allow formation of a continuous
strand consisting of a solid shell encapsulating liquid material, cooling the
strand by spraying water while orienting the mould to horizontal position
for complete solidification till the solidified shell cut into discrete slabs, the
method comprising the steps of:
- advance prediction of tear, crack, break, or lumb formation leading
to break-out, in the formative shell in the mould at the narrow
faces of the emergent strand; and
- preventing such break-out by raising alarm sufficiently in advance
enabling corresponding reduction in casting speed, wherein:-
- multi-parametric analysis of the casting process is conducted in
real-time through an Artificial Neural Network (ANN) to generate a
value of abnormality index (AI) representing the deviation from
normal process conditions;

- a short-term-time or moving average (MAI) of the abnormality
index (AI) is compared against a long-term-time average of the
MAI (NICC) that represents the normal casting; and
- a sharp increase in the short-term-time AI (MAI) over the long-
term-time AI (NICC) is predicted as an on-set of lumb-type break-
out.
2. The method as claimed in claim 1, wherein a plurality of known process
parameters acquired in real-time and input to the AAN to generate AI,
comprise:
absolute value of time derivatives of levels summed over previous
thirty steps;
at most two wide face heat fluxes;
absolute value of difference said two wide face heat fluxes;
signed difference between wide face average and narrow face
maximum heat fluxes;
average of two friction values; and
absolute value of time derivatives of average friction.

3. The method as claimed in claim 1 or claim 2, wherein a threshold value of
ratio of MAI and NICC as defined, and compared on-line, and wherein a
minimum threshold value of MAI is separately defined.
4. The method as claimed in claim 1 or 3, wherein when both the predefined
threshold value of MAI, and the predefined threshold ratio of MAI and
NICC exceeds, a potential alarm is raised.
5. The method as claimed in claim 1 or 4, wherein the potential alarm is
converted to actual alarm after allowing the former to continue for a
reaction period for example, 10 seconds.
6. An on-line system for advance-prediction of "lumb" type of break-outs to
initiate preventive measures in a continuous casting process of steel
manufacturing in particular using thin-type slab casters, comprising an
Artificial Neural Network (ANN) structurally formed of at least three layers
of neurons, a first layer and a second layer respectively being an input
and output layer, the balance being hidden layers, the neuron in each
layer summing up all inputs from the neuron of the preceding layers, and
multiplying by associated weight between the two neurons to provide a

non-linear transform to the sum to generate an output which constitutes
the input for the neurons of the next layer; and a plurality of sensors and
a data acquisition means acquiring in real-time the data relating to
multiple process parameters of the casting process; and a processor to
receive data from the sensors and input processed data to the ANN.

ABSTRACT

An online method for advance-prediction of "lumb" type break-outs to initiate
preventive measures in a continuous casting process of steel manufacturing
using thin-type slab caster, the process comprising the steps of passing the
molten steel through a water-cooled substantially vertically-aligned lubricated
mould to allow formation of a continuous strand consisting of a solid shell
encapsulating liquid material, cooling the strand by spraying water while
orienting the mould to horizontal position for complete solidification till the
solidified shell cut into discrete slabs, the method comprising the steps of
advance prediction of tear, crack, break, lump or lumb formation leading to
break-out, in the formative shell in the mould at the narrow faces of the
emergent strand; and preventing such break-out by raising alarm sufficiently in
advance enabling corresponding reduction in casting speed, wherein, multi-
parametric analysis of the casting process is conducted in real-time through an
Artificial Neural Network (ANN) to generate a value of abnormality index (AI)
representing the deviation from normal process conditions; a short-term-time or
moving average (MAI) of the abnormality index (AI) is compared against a long-
term-time average of the MAI (NICC) that represents the normal casting; and a
sharp increase in the short term-time AI (MAI) over the long-term-time AI
(NICC) is predicted as an on-set of lumb-type break-out.