FIELD OF THE INVENTION
The present invention relates to a system involving hybrid model based control for
predicting the entry temperature for finish roiling stand based on input data relating to
temperature at the exit of roughing stand. More particularly, the present invention relates to
a system involving mathematical-statistical-Artificial Neural Network (ANN) based hybrid
thermal model for delay table for online acquisition, analysis, control and display of process
data, independent of operators' expertise/experience, to ascertain/forecast the data relating
to temperature profile along the rolled hot transfer strip length after it is passed in the delay
table of a hot strip mill having coil box and crop shear, in order to determine and display the
finish rolling temperature to favor setting of roll gaps at finish rolling stands. The system
software is adapted to receive online input data from a pyrometer located after the last
roughing stand via a PLC system connected with OPC network and a computer system called
the Process Work Station (PWS). Importantly, the model software is adapted for accurate
temperature prediction wherein mathematical model has been derived from Stefan-
Boltzman's law of thermal radiation and the statistical model is a linear regression model.
Also the model software enable the system to be adapted for on-line automatic calibration
of hybrid model for training/weights for different steel groups. The output is displayed to
operator in another computer system called Operator Workstation, facilitating the operator
controls the finishing stand settings/parameters including roll gap based on the feedback
received from the system including finish stand entry temperature. The system thus favor
determining the settings of finish stand parameters in a reliable and fool proof manner
eliminating uncertainties involved in estimation based on experience while the roughing mill
output data and delay table length are given/known.
BACKGROUND OF THE INVENTION
It is known in the existing art that strips of different thickness and width are rolled from slab
in a Hot Strip Mill. One of the Hot Strip Mill Plant of the applicants comprised 3 Roughing
stands and 6 Finishing stands. The first roughing stand (R0/V0) is a combination of a
horizontal stand and a vertical stand. The other two roughing stands (R1 and R2) are 4 high
horizontal stands. There is a delay table after R2 stand, one coil box and a crop shear at the
end of the delay table. There are six numbers of 4 high finishing stands (F1 to F6) and two
hydraulic down coilers. In conventional practice, the operator at the finishing stand

estimated/predicted the first finishing stand entry temperature based on experience and
sets the roll gap based on the measured temperature at the exit of second roughing stand
R2 such that desired finish rolling is carried out within required temperature range. This
method of prediction of finish roll temperature thus involved inaccuracy and uncertainty to a
large extent to thereby resulting variation on the rolling parameters and affecting adversely
the end quality of finish rolled steel product.
There has thus been a continuing need in the rolling process in hot strip mill comprising the
roughing stand and finishing stand to developing an online computer based system for
measurement, control as well as display of the temperatures at the required stage/zone
along the strip in order to ascertain the parameter settings of the finishing stand in a much
deterministic manner without reliance on the operators experience/expertise in order to
eliminate the inaccuracy as well as undesired variations and carry out the finish rolling in
controlled manner resulting in superior and consistent quality of the end product.
Advantageously, the system of the invention would involve a mathematical-Statistical-
Artificial Neural Network (ANN) based hybrid thermal model for accurate prediction of
temperature after delay table adapted to provide a temperature profile along the strip and
display means to the operator through PLC-Server system for attaining desired control on
the finish rolling process parameters.
OBJECTS OF THE INVENTION
The basic object of the present invention is thus directed to developing a system for
accurate prediction of finish stand entry temperature after delay table adapted to provide a
temperature profile along the strip and display means to the operator through PLC-Server
system connected through OPC network, facilitating the operator to set finish rolling
temperature/parameters in a reliable manner.
Another object of the present invention is directed to providing a computer system involving
a mathematical-statistical-Artificial Neural Network (ANN) based hybrid thermal model for
delay table for desired computation of finishing entry temperature based on roughing stand
exit temperature using said hybrid thermal model, favoring determination of the roll gap by
operator.

A still further object of the present invention is directed to developing hybrid thermal model
based software enabled computer system wherein mathematical model has been derived
from Stefan-Boltzman's law whereas the statistical model is linear regression model.
A further object of the present invention is directed to developing a software enabled
computer system wherein said hybrid thermal model based software is loaded on Process
Workstation that works on continuous basis so that the model output is shown to the
operator.
A still further object of the present invention is directed to developing a software enabled
computer system adapted to take online input data from a pyrometer located after the last
roughing stand through a PLC system and computer systems called Process Work Station
(PWS).
A still further object of the present invention is directed to developing a software enabled
computer system adapted to provide output feedback displayed to operator in another
computer system called Operator Workstation(OWS).
A still further object of the present invention is directed to developing a software enabled
computer system wherein the PWS/OWS system has been connected with the PLC system
using OPC network software.
A still further object of the present invention is directed to developing a software enabled
computer system wherein the hardware arrangement comprising the servers, workstations,
PLC system and the sensor/actuators that are deployed in three different levels for
implementation of the software.
A still further object of the present invention is directed to developing a software enabled
computer system advantageously providing required information displayed to the operator
to thereby ensure control of the finishing stand settings based on the feedback data.

SUMMARY OF THE INVENTION
The basic aspect of the present invention is thus directed to a system for prediction of
temperature profile along strip length after it passes in the delay table of a hot strip mill
comprising:
means for acquiring online input data after the last roughing stand ;
computer means adapted for predicting the said temperature profile based on said input
data and output derived based on mathematical-Statistical-Artificial Neural Network (ANN)
based hybrid model and generating a feedback output;
means for controlling the finishing strand settings based on said feedback output.
Another aspect of the present invention is directed to a system wherein said computer
means comprises a process work station and said feedback output is displayed in a separate
operator work station for the operator to control the finishing strand settings based thereon.
A further aspect of the present invention is directed to a system wherein the said hybrid
model is adapted for accurate prediction of temperature after delay table and display the
same to operator through a PLC-server system.
A still further aspect of the present invention is directed to a system comprising OPC
networked PLC -server system.
According to yet another aspect of the present invention directed to said system comprising
hot strip mill comprising 3 roughing stands and 6 finishing stands, the first roughing
stand(R0/V0) being a combination horizontal stand and a vertical stand, the other two

roughing stands (R1 and R2) being 4 high horizontal stands,a delay table after R2 stand ,
one coil box and a crop shear at the end of the delay table, six numbers of 4 high finishing
stands (Fl to F6) and two hydraulic down coilers;
process work station adapted to execute the said hybrid model based prediction ;
operator work station adapted to display model output to the operator; and
PLC based system operatively connected involving OPC network ,
wherein the temperature after R2 is provided to the process work station and in the said
process work station the R2 temperature is stored in a MS Access database , said
mathematical-Statistical-Artificial Neural Network (ANN) based hybrid model adapted to
read said temperature data from the database and generate Fl temperature with the
output data passing through the PLC system and output displayed in the operators work
station.
A still further aspect of the present invention directed to said system wherein the said
mathematical-Statistical-Artificial Neural Network (ANN) based hybrid model is based on
Stefen-Boltzman's law of radiation and the following equation:
Temperature drop due to radiation heat loss given by,

where TR2 = Temp, of hot transfer bar after R2, 1= length of delay table = 65m, vd= speed
of bar delay table, tCB= time delay in coil box. tCB= 0, if coil box is not used, tcs=time delay
of cutting in crop shear. Kr = Coefficient of radiation heat transfer loss. Kr, a factor which is
dependent upon the view factor of radiation has been considered as a mathematical
calibration factor in the model.

A still further aspect of the present invention is directed to said system comprising a linear
regression based statistical model, the slope and bias of the regression equation being
considered as model calibration factors, the output of mathematical and statistical models
along with the R2 temperature are input to the feed-forward ANN model with back-
propagation algorithm for training, the weights and biases of ANN model is also considered
as model calibration factors, said system adapted for on-line automatic calibration of hybrid
model for different steel groups.
A still further aspect of the present invention is directed to said system comprising
hierarchical input data structure for calibration purposes, selective initial ANN training
weights and input data sampling done on average basis and normalization of inputs data
segments based on linear interpolation.
The present invention and its objects and advantages are described are in greater details
with reference to the following non limiting illustrative figures.
BRIEF DESCRIPTION OF THE ACCOMPANYING FIGURES
Figure 1: is the schematic illustration of the existing layout arrangement of the roughing
and finishing stands with delay table, coilbox and cropping shear located inbetween,
wherein the hybrid system of the present invention is applied for online process control.
Figure 2: is the schematic illustration of the computer system according to the present
invention involving the hardware deployment in three separate levels comprising the PWS/
OWS/servers, PLC system and the sensor/actuator/pyrometers for implementation of the
software.
Figure 3: is the illustration of the screenshot of the model output at operators end showing
the F1 temperature prediction at entry of first finish stand based on the measured
normalized R2 temperature at the exit of last roughing stand and the computed model error
(%).

Figure 4: is the schematic illustration of the conceptual diagram of the mathematical-
statistical hybrid model for computation of temperature after delay table.
Figure 5: is the schematic illustration of the hierarchy of different steel data group
comprising a grade, a target width, a target thickness and a coilbox use.
Figure 6: is the schematic illustration of the selection of the initial training weights of ANN
instead of random weights.
Figure 7: is the schematic illustration of the online automatic calibration/training module of
hybrid model for different steel groups.
DETAILED DESCRIPTION OF THE INVENTION WITH REFERENCE TO THE
ACCOMPANYING FIGURES
The present invention is directed to developing an online software enabled system for
accurate prediction of temperature after delay table and providing temperature profile along
the strip length in hot strip mill involving a mathematical-Statistical-Artificial Neural Network
(ANN) based hybrid thermal model and display of the system output to operator through
PLC-Server system in order to predict temperature at entry of finishing stand as well as
specifically set the finish rolling parameters including the roll gaps. The system is adapted to
receive online input data from a pyrometer located after the last roughing stand through a
Siemens make PLC system and a computer system called Process Work Station (PWS). The
output is displayed to operator in another computer system called Operator Workstation.
Operator controls the finishing stand setting based on this feedback.
Reference is first invited to the accompanying Figure 1, that schematically illustrates
existing hot strip mill facility in one of the steel plants of the applicant that comprised three
roughing stands (RS) and six numbers horizontal finish rolling stands. There is delay table
(DT) of specified length and a coil box (CB) in between the roughing stand and the finishing
stand. It is clearly apparent from the Figure 1 that the first roughing stand (R0/V0) is a
combination of horizontal stand and a vertical stand, whereas the other two roughing stands

e.g. R1 and R2 next in a row after Ro/V0, are 4 high horizontal stands. The delay table (DT)
is located after R2 stand. There are also one coil box (CB) and a crop shear (CS) at the end
of the delay table. There are six numbers of 4 high finishing stands numbered as F1 at entry
to F6 at the exit of finish rolling stage and two hydraulic down coilers. In the existing
practice the operator of finishing stand sets roll gap on the basis of temperature of first
finishing stand entry temperature. Also conventionally, the operator predict finishing entry
temperature from R2 measured temperature on the basis of his experience and thus
involving inaccuracy and error in determining such temperature.
The present invention is directed to a computer based system installed in the hot strip mill
comprising a process work station (PWS) in which the mathematical-statistical-ANN based
hybrid thermal model is loaded that work on continuous basis. An Operator Work Station
(OWS) shows model output to the operator. This system has been connected with the PLC
system using OPC network software. Accompanying Figure 2 schematically illustrate the
deployment of the hardware of the PLC-Sen/er based computer system in different levels.
The system records and input the temperature next to R2 to the PWS, such that in PWS, the
R2 temperature is stored in a MS Access database. The software involving the invented
hybrid thermal model reads the temperature data from the database and calculates Fl
temperature. The output data passes through the PLC system in similar fashion so that the
output is shown to the operator. It is seen from the Figure 2, that the system comprises
OWS and PWS alongwith the PDI server in level 2 which are integrated with PLC and level l
server while the PLC is coupled with the sensors and actuators (Pyrometers) to establish
bothway communication for input and processed data with the level 0 at the operating level
at delay table and finishing stand of hot strip mill, so that the above hardware arrangement
favours implementation of the hybrid software based software for desired controlled setting
of finish roll parameters.
As already described, the present invention works on a hybrid thermal model based
software to operatively integrate the hardware elements at different levels.
The hybrid thermal model consisting of mathematical, statistical and artificial neural
network based hybridisation wherein (i) the mathematical model has been derived from
Stefan-Boltzman's law, whereas (ii) the statistical model is a linear regression model.

The mathematical model is based on Stefen-Boltzman's law of radiation which has been
further simplified into following equation:
Temperature drop due to radiation heat loss is given by,

where TR2 = Temperature of hot transfer bar in °C measured after R2; I= length of delay
table = 65m; vd= speed of bar on delay table; tcB= time delay in coil box, tCB= 0, if coil box
is not used; tCS=time delay of cutting in crop shear; Kr = Coefficient of radiation heat
transfer loss. Kr, is a factor which is dependent upon the view factor of radiation that is
considered as a mathematical calibration factor in the invented model. A linear regression
based statistical model has been developed. The slope and bias of the regression equation
have been considered as model calibration factors. The output of mathematical and
statistical models along with the R2 temperature are input to the feed forward ANN model
with back-propagation algorithm for training. The weights and biases of ANN model is also
considered as model calibration factors. A software has been developed for on-line
automatic calibration of hybrid model for different steel groups wherein a steel group
consists of a grade, a target width, a target thickness and a coilbox use. The model in
operation is automatically calibrated for different steel groups. The screen shot of the hybrid
model output is schematically illustrated in the accompanying Figure 3 showing the stand
wise predicted temperatures in finish stands with target thickness of 2.3mm and target
width of 1055mm. The screen shot of Figure 3 shows the prediction temperature at entry of
Fl based on normalized R2 temperature input to the system. The model also compute the
error % for each data set.
The working of the hybrid thermal model for delay table to define temperature for the strip
length and accurately predict the Fl entry temperature based on measured temperature at
exit of F2 is implemented by way of the hardware configuration and the software
implementation as stated above, comprising the steps of
(a) Development of mathematical, statistical and artificial neural network based hybrid
thermal model for. delay table wherein the mathematical model has been derived

from Stefan-Boltzman's law whereas the statistical model is a linear regression
model.
(b) Identification of model calibration factors for each model component.
(c) Development of data communication system among pyrometers, PLCs, PWS and
OWS for segment wise collection of data and segment-wise prediction of
temperature of coils.
(d) Designing database structure for Primary datainputs of coil, Pyrometer
temperatures, coilbox status and model calibration factors.
(e) Developing the software for on-line automatic calibration/training of hybrid model
for different steel groups, each group characterized by a grade, a target width, a
target thickness and a coilbox use.
(f) Assigning initial weights of ANN by selection instead of random weights, applied on
a data set to determine/predict the temperature after delay table according to the
thermal model. When there is no training data, the output becomes average of
mathematical and statistical model.
(g) The software collects temperature data at every 0.5 minute interval and averages
10 data to generate 1 sampling point.
(h) Normalization of sample data is carried out by linear interpolation method.
(i) On-line trial of the model showed that the output of calibrated model predicts
temperature (F1-F6) with an error of less than 1.5%.
Reference is now invited to the accompanying Figure 4 that schematically illustrate the
conceptual diagram for the above described hybrid thermal model for operation of the
model on the input data set for different steel group. Accompanying Figure 5 shows the
hierarchy of data group used for computing the finish temperature by the hybrid model.
Reference is also invited to the accompanying Figure 6 that shows the use of selected
initial weights 1.0, 0.5 and 0.0 for operation on a data segment to predict the temperature
after delay table.
Reference is now invited to the accompanying Figure 7 that schematically illustrate the
online automatic calibration/training module of hybrid model for on-line data exchange for
different steel groups.
It is apparent from the system architecture and data communication/exchange in the
developed hybrid system, the data extraction program in PLC system interfaced with a level

1 server for processing; it collects temperature data at every half minute interval on
continuous mode. The computer system of the invention comprising the Process Work
Station in which the hybrid model is loaded and work on continuous basis and OWS that
shows the output to the operator. The PWS and the OWS are at level 2. This system is
connected with the PLC system using OPC network software for data communication/
exchanged. The average is computed from 10 data (temperature) to form one sample point.
Also the normalization of sample input data segments is performed on the basis of linear
interpolation. The model software developed is used to integrate the hardware elements at
different levels to implement the steps of the hybrid thermal model comprising the
mathematical, statistical and ANN for prediction of temperature precisely along the length of
transfer bar after delay table. A hierarchical input data structure has been developed for
calibration purpose. The software also implements a method of selection of initial ANN
training weights to compute/predict the temperature at finish stand entry and a
temperature profile along strip length after delay table. The model software reads the
temperature data from the data base and calculate F1 temperature. The mathematical
model finds out the temperature drop due to radiation heat loss given by a simplified
empirical formula based on Stefen Boltzman law as given in equation (1) already stated and
a linear regression based statistical model is developed. The view factor of radiation is
considered as a mathematical calibration factor in the model and also the slope and bias of
the regression equation are considered as other model calibration factors. The output of
mathematical and statistical models along with the R2 temperature are input to the feed-
forward ANN model with back-propagation algorithm for training. The weights and biases of
ANN model is also considered as model calibration factors. The developed software is
adapted to on-line automatic calibration of hybrid model for different steel groups
comprising Grade, Coil width, Coil thickness, Coil box status and segment number for
segment wise hierarchical processing of input data structure for calibration purpose as well
as to compute the Fl temperature corresponding to data set for R2 temperature and speed.
Initial training weights of ANN have been selected as 1.0, 0.5 and 0.0 instead of random
weights. When there is no training data the output become average of mathematical and
statistical model.
It is thus possible by way of the present invention to developing an online computer system
and method for accurate and precise prediction of the entry temperature of finish stand and
temperature profile along strip length after delay table, based on input data relating to
roughing stand exit temperature and speed by involving a mathematical, statistical and

WE CLAIM:
1. A system for prediction of temperature profile along strip length after it passes in the
delay table of a hot strip mill comprising:
means for acquiring online input data after the last roughing stand ;
computer means adapted for predicting the said temperature profile based on said input
data and output derived based on mathematical-Statistical-Artificial Neural Network (ANN)
based hybrid model and generating a feedback output ;
means for controlling the finishing strand settings based on said feedback output.
2. A system as claimed in claim 1 wherein said computer means comprises a process work
station and said feedback output is displayed in a separate operator work station for the
operator to control the finishing strand settings based thereon.
3. A system as claimed in anyone of claims 1 or 2 wherein the said hybrid model is adapted
for accurate prediction of temperature after delay table and display the same to operator
through a PLC-server system.
4. A system as claimed in anyone of claims 1 to 3 comprising OPC networked PLC-server
system.
5. A system as claimed in anyone of claims 1 to 4 comprising
hot strip mill comprising 3 roughing stands and 6 finishing stands, the first roughing
stand(R0/V0) being a combination horizontal stand and a vertical stand, the other two
roughing stands (R1 and R2) being 4 high horizontal stands, a delay table after R2 stand ,
one coil box and a crop shear at the end of the delay table, six numbers of 4 high finishing
stands (F1 to F6) and two hydraulic down coilers;
process work station(PWS) adapted to execute the said hybrid model based prediction ;

operator work station(OWS) adapted to display model output to the operator; and
PLC based system operatively connected involving OPC network ,
wherein the temperature after R2 is provided to the process work station and in the said
process work station the R2 temperature is stored in a MS Access database , said
mathematical-Statistical-Artificial Neural Network (ANN) based hybrid model adapted to
read said temperature data from the database and generate F1 temperature with the
output data passing through the PLC system and output displayed in the operators work
station.
6. A system as claimed in anyone of claims 1 to 5 wherein the said mathematical-Statistical-
Artificial Neural Network (ANN) based hybrid model is based on Stefen-Boltzman's law of
radiation and the following equation:
Temperature drop due to radiation heat loss given by,

where TR2 = Temp, of hot transfer bar after R2, I= length of delay table = 65m, vd= speed of bar
delay table, tCB= time delay in coil box. tCB= 0, if coil box is not used, tCS=time delay of cutting in crop
shear. Kr = Coefficient of radiation heat transfer loss. Kr, a factor which is dependent upon the view
factor of radiation has been considered as a mathematical calibration factor in the model.
7. A system as claimed in claim 6 comprising a linear regression based statistical model, the
slope and bias of the regression equation being considered as model calibration factors, the
output of mathematical and statistical models along with the R2 temperature are input to
the feed-forward ANN model with back-propagation algorithm for training, the weights and
biases of ANN model is also considered as model calibration factors, said system adapted
for on-line automatic calibration of hybrid model for different steel groups.

8. A system as claimed in anyone of claims 1 to 7 comprising hierarchical input data
structure for calibration purposes, selective initial ANN training weights and input data
sampling done on average basis and normalization of inputs data segments based on linear
interpolation.
9. A system for prediction of temperature profile along strip length after it passes in the
delay table of a hot strip mill substantially as hereindescribed and illustrated with reference
to the accompanying figures.

The invention relates to a system for predicting accurately the finishing entry temperature
based on input temperature of last roughing stand in HSM, involving mathematical-
Statistical-Artificial Neural Network (ANN) based hybrid thermal model to determine
temperature profile along length of hot rolled transfer strip after the delay table. The system
software is adapted to receive online input data from a pyrometer to process work station
(PWS) through a PLC system connected via a OPC network. Importantly, the system uses
mathematical model derived from Stefan-Boltzman's law whereas the statistical model is a
linear regression model. Also the system is adapted for on-line automatic calibration of
hybrid mode! for training/weights for different steel groups. The output is displayed to
operator on Operator Workstation(OWS) via PLC interface. Operator controls the finishing
stand settings/parameters based on the feedback received from the system. The system
thus favor determining the settings of finish stand parameters in a reliable and fool proof
manner eliminating inaccuracies in estimation based on operator's experience.