FIELD OF THE INVENTION
The present invention relates to a system for optimum design of hot rolled steel
products using cyclic evolutionary and classical algorithms. The present invention
can be utilized in any hot strip mill for rolling a new product with a target mechanical
properties such as yield strength (YS), ultimate tensile strength(UTS) and %
elongation(%el). This system is designed such that without considerations of the
metallurgical phenomena during hot rolling, a new product can be made. This
system automatically co-ordinates the implicit parameters of material composition,
geometry and processing conditions to design a product with desired material
properties using artificial intelligence based models. The system is a hybridization of
artificial neural network, evolutionary algorithm and classical algorithm by
synchronizing the advantageous features of both of these methods.
BACKGROUND OF THE INVENTION
Hot-rolling is done with an aim to change the geometry of the slabs to meet the
dimension of the final product. This is done by reheating slabs to the desired
temperature required for homogenization of chemistry and dissolution of precipitates
formed during casting. The material is then rolled in the finish rolling mill with a
typical combination of finish rolling (FRT) and coiling temperatures (CT). After finish
rolling the coils are cooled on run out table to achieve the target coiling temperature
by following a particular spray pattern to achieve the desired cooling rate. The
precise control of reheating temperature, FRT, CT and the cooling rate is the key
behind getting right combination of microstructure and precipitation which helps in
achieving the target mechanical properties. The detail understanding of evolution of
the austenite microstructure during hot rolling plays an important role in converting
carbon steel into sophisticated high strength low alloy (HSLA) steel.
Optimization of chemistry and process parameters plays a big role in designing a
new hot rolled steel product without compromising the properties. Products with
desired mechanical properties can be very useful in steel industry and will have vast

impact on its application domains. Currently the design of chemistry for new product
development is totally based on the past experience and trial-and-error method. As
there is no pilot plant, minimum 150 Mt heat is made to take a trial which is very
costly and time consuming.
The final mechanical property of the hot rolled steel is the outcome of geometrical
parameters, chemical parameters and process parameters. Many kinds of grades of
steel with different combination of chemical compositions like micro-alloyed grades
and plain C-Mn grades steels are rolled with different operating parameters to
achieve different properties. Sometimes during hot rolling process, effort is made to
take care of one property at the expense of other properties. Due to the emerging
applications of steels, the production of hot rolled sheets has been improved over
the period of time. Because of high end applications, steel sheets require very rigid
control of the conflicting mechanical properties keeping cost and mill capability in
mind.
However optimizing the properties in microallyed grades, particularly when used in
combination, is difficult. It is because of the fact that the best properties in micro-
alloyed steels are obtained at the right combination of reheating temperature,
amount of reduction in various rolling stands, finish rolling and coiling temperatures
and the cooling rate between finish rolling and coiling. Needless to say that the
optimum value for all these would differ with varying amount of alloying addition
and mill configuration. This why any first principle based models would be difficult
for designing the correct processing parameters and chemistry for microallying
grades in order to get the best combination of properties. The microalloyeds steel
with correct combination of mechanical properties can be suitable alternative for low
strength variety, which would reduce the weight of the components and the
consumption. This would be helpful in reducing the overall carbon foot-print.
However the processing conditions in manufacturing are complex and as a result its

impact in combination with the chemistry on the mechanical properties is difficult to
model.That’s why the need for such a system came into place.
OBJECTS OF THE INVENTION
An object of the invention is to propose a system for optimum design of hot rolled
steel products using cyclic evolutionary and classical algorithms.
Another object of the invention is to collect different data from different sensors
located at various locations in a hot strip mill of a steel plant.
A still another object of the invention is validation of data to configure a committee
of neural network based models for prediction of mechanical properties of hot rolled
steels with more than 95% accuracy.
A further object of the invention is to propose a hybrid algorithm for adaptation in
the system which utilizes both dominance and decomposition (MOEA/DD) concepts
including normalized normal constraint (NNC) to optimize several objective
problems (MaOP’s). Here both unconstrained and constrained forms are used.
SUMMARY OF THE INVENTION:
Accordingly, there is provided a system for optimum design of hot rolled steel
products using cyclic evolutionary and classical algorithms. This system is able to
recommend for rolling the steel with optimized sets of chemistry and processing
parameters in order to achieve desired mechanical properties of hot rolled steels.The
aim here is to optimize several conflicting mechanical properties within the given
range of parameters by using different optimization algorithms (evolutionary as well
as classical). This system is able to handle more than three objectives at a time in a
constrained environment which is imposed by user in plant. The invention further
provides, a novel hybrid methodology i.e. searching the local optimum in the global

basin and a cyclic combination of MOEA/DD and NNC and I which is processed in
two steps. In the first step, an approximated Pareto front (Pareto front I) is
constructed using MOEA/DD. The improved Pareto front (Pareto front II) is
approximated using classical gradient-based technique (NNC) in the second step.
BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWINGS:
Fig.1.a: Architecture of the system of the invention
Fig.1.b: Front end of the system of Fig.1.a
Fig.2: Information flow of the system of Fig.1.a
Fig.3: Information of the module 108 of the system for calculation of
properties
Fig.4: Information Flow of the optimizer 112 of the system
Fig.5: Information flow of NNC module
Fig.6: Information flow of mating module section
Fig.7: Information flow of binary tournament module
Fig.8: GUI for entering target variable values
Fig.9: GUI for entering constraints values
Fig.10: GUI to observe parameter values

DETAILED DESCRIPTION OF THE INVENTION
It is known that all over the world steel industries cause a lot of greenhouse gas
emission. One way to counter the phenomenon is to reduce the consumption of
steel by downsizing the component thickness and eventually the weight of the
components. This is true not only for automobile industries but also for steels being
used in construction. The best way to do this is to reduce the production of low
strength C-Mn steels and replace them with higher strength microalloyed steel. The
aim of this invention is to develop a methodology for designing such desired grades
in steel plants which enables optimisation of the alloy addition to reduce the cost
and still achieve best properties.
The optimisation of microallyed becomes interesting because it manifests
complimentary strengthening mechanisms, specifically grain refinement and
precipitation hardening. In the optimisation model improved properties have been
targeted considering the contribution of each strengthening mechanism such as solid
solution hardening, grain refinement, precipitation hardening and dislocation
hardening. Precipitation hardening increases strength but sometimes contributes to
brittleness. Grain refinement increases strength and at the same time also improves
toughness. Sometimes it is also necessary to design an alloy to accommodate the
final range of mechanical properties keeping cost and mill capability in mind.
In real world applications, simultaneous satisfaction of multiple numbers of
conflicting properties are required to be addressed that greatly influence the
performance, geometric specification of a system leading to the development of a
multi-objective optimization problem (MOP).
With respect to the optimization of the mechanical properties of the hot rolling mill,
several conflicting mechanical properties which can be used as different objectives
are yield strength (YS), ultimate tensile strength (UTS) and % elongation of hot
rolled steel. Parameters related to the chemistry and processing are going to be

used as decision variables. There will be certain bounds or constraints on the
objectives. As such there are no other explicit constraints for this problem.
As the property of the steel depends on its microstructure, which is a complex
function of composition and processing parameters, a good model connecting the
properties and the parameters is very useful for improving the properties of the
steel. In this case artificial neural network (ANN) based models are used to establish
the nonlinear correlation between the composition and process variables with the
properties of the steel. It is very much necessary to find the most suitable
combination of chemistry and process parameters through optimization which
simultaneously satisfy several conflicting desired mechanical properties of steel. So it
is interesting to use ANN model to establish the complex nature of strength
evolution in hot rolled material.
Since objectives in MOP are of conflicting nature with one another, no single optimal
solution exists to optimize all the objectives simultaneously. Besides, a set of trade-
off solutions on the Pareto Optimal (PO) front is obtained which can help the
decision maker to understand the conflicting nature of different objectives and
choose the preferred solution as per the present requirement. Generally, a multi-
objective optimization problem can be solved by two approaches, i.e. the first one
takes help of population-based or metaheuristic evolutionary algorithms and the
second one utilizes the gradient-based classical techniques. Both of them
(approaches) have shown their efficacy in solving the MOPs. Usually, the final Pareto
front is approximated in a single simulation run in the population-based multi-
objective evolutionary algorithms (MOEA’s). In order to obtain a good distribution of
solutions in the Pareto front, MOEAs are generally designed to achieve a balance
between convergence and diversity, respectively. Convergence is the measure of
how close the obtained Pareto solutions from the true solutions, which should be as
small as possible, whereas, a uniform spread of solutions in the Pareto front is
measured as diversity.
In this work, an artificial neural network based model has been developed to predict
the mechanical properties of steel. ANN model developed in this work was also used

as the cost function for optimising the properties of micro-alloyed steel. The
generated pareto fronts give a wide range of combination of strength and elongation
which helps in choosing the right design of chemistry and processing parameters for
achieving the desired property. Both evolutionary and classical techniques are
utilized to optimize the chemistry and processing parameters in order to achieve the
desired mechanical properties of hot rolled steel mechanism. In a population-based
approach, multi-objective optimization based on decomposition and dominance
(MOEA/DD) is used and Normalized normal constraint (NNC) is used in the classical
approach category. Both of them are applied to a case study of hot rolled steel
mechanism with an aim of maximizing yield strength (YS), ultimate tensile strength
(UTS) and percent elongation (% elongation). Moreover, the evolutionary algorithm
has a tendency to provide global (near global) basin solutions whereas classical
techniques are efficient in providing local optima. Utilizing the merits of both the
approaches, a novel hybrid methodology has been proposed. In the hybrid
approach, the search power of the evolutionary algorithms to find the global basin is
merged with the ability of the classical approaches to find a local optimum (i.e. this
is hopefully searching the local optimum in the global basin) which can be hopefully
lead to a global optimum. Application of hybrid approach is also applied to the case
study of hot rolled steel mechanism.
Few graphical user interface of the system is given in Fig 8,9,10 for the user to
operate.

WE CLAIM:
1. A system for optimum design of hot rolled steel products using cyclic
evolutionary and classical algorithms, the designed product being produced in
a hot strip mill (100) in a steel plant, the system comprising:
a plurality of sensors (S1, S2, S3………….S10), placed at various locations in
the hot strip mill, the plurality of sensors being configured to sense values
corresponding to parameters of slab drop out temperature), soaking
time),slab retention time , roughing mill exit temperature, finish rolling
temperature, coiling temperature, thickness, last stand speed in finishing mill
and composition of liquid steel and forward the sensed values to a data
storage means (104);
the data storage means (104) being coupled to the plurality of sensors (S1,
S2, S3………….S10), and configured to store and forward the sensed data to a
calculating means (108);
the calculating means (108) being coupled to the data storage means (104),
and configured to receive and feed the sensed data in a module including in
an equation

Where in Xi is the value received from ith sensor ,wi and θi are the
coefficients of parameter of ith sensor and wj and θ are coefficients of hj,
determining the values of YS, UTS and %el and forwarding the values of
“YS”, “UTS” and “%el” to an optimizer (112), and forward the out putted
optimized process parameters being forwarded to a decision making means
(116);

the decision making means (116) being coupled to the calculating means
(108), and configured to receive the optimized process parameters to judge
the output of “YS”, “UTS” and “%el”, the judged values being forwarded to a
user interface (118), the judgement being based on following assumptions:
if lower limit <=YS