Abstract The primary objective of steelmaking through
Basic Oxygen Furnace (BOF) process is to achieve desired
end point carbon content, temperature and percentage
composition at the lowest cost and in the shortest possible
time. As of now, most widely used models for prediction of
parameters of converter steelmaking are mechanistic model,
statistical model and neural network model for the
prediction of the end point carbon content and temperature
from BOF process parameters with reasonable accuracy.
The (BOF) process is a widely preferred and effective
steelmaking process due to its higher productivity and low
production cost. The process of converter steel making is
complicated and not completely understood as it involves
multiphase physical chemical reaction at high temperature.
Obtaining molten steel of desired chemical composition is
the objective of the process. Obviously, in the converter steel
making , the end point carbon content and temperature of
the molten steel are important controlling parameters to
ascertain whether the molten steel of desired quality is
achieved or not.
In the present paper, the authors have made an attempt
to develop model for end point carbon and temperature with
the latest methodology i.e., Adaptive %eural Fuzzy Inference
System (A%FIS) and then have brought out the comparison
of the results achieved in %eural %etwork and GR%%
models. Results from A%FIS model predict more accurately
in contrast to those from BP%% model vis+à+vis the
measured carbon content and temperature.
Keywords BOF, End point temperature, End point
carbon, predictive model, A%FIS, GR%%, BP%%.
I. INTRODUCTION
BOF steelmaking is a very important during steel
production process. The process is very complex due to
harsh environment and high temperature during the steel
making process. The endpoint carbon content and
temperature of the molten steel are the parameters to be
checked up whether the molten steel is of desirable
quality or not. The carbon content and temperature are the
main controls in steel melting process. The widely used
modelling approaches for BOF process include
mechanistic models, statistical models while neural
network models to a certain extent. The existing control
models based on oxygen balance and thermal equilibrium
theory are achieved on the condition of certain
hypotheses, so it can not reflect the relationship between
the components produced and endpoint carbon content
and temperature for steel making. With these methods,
more than 50% of the heats have to be reblown to
achieve the desired end conditions. This involves loss of
production and extra raw material cost. Kubat et al. [1]
proposed a fuzzy modelling approach for the control of
BOF process. As a result of the application of the
proposed modeling, acceptable levels of compatibility
were achieved compared to the empirical BOF data in an
integrated steel plant at Turkey. Bigeev and Baitman [2]
adopted different intelligent method to describe the BOF
steelmaking process. Although artificial intelligence
model give a better result than conventional methods, it
limits the model precision due to high dimensionality.
Szekely [3] discussed the necessity to simplify the input
variables to reduce the complexity and improve the
generalization capability of the model. At present,
adaptive neuro fuzzy modeling technique has been used in
BOF steelmaking process to establish model to predict
end point carbon content of metal from important process
variables.
Fuzzy logic controllers (FLC) yield superior and
faster results, without the use of accurate mathematical
model of the plant and work well for complex nonlinear
multidimensional system. FLC are based on the fuzzy
set and fuzzy logic theory originally advocated by Lotfi
A. Zadeh [4]. The fuzzy control is adaptive in nature and
gives robust performance. The main design problem lies
in the determination of the consistent and complete rule
set and the shape of the membership functions. A lot of
modifications, trial and error have to be carried out to
obtain the desired response, which is time consuming.
NeuroFuzzy software tools work as an intelligent
assistant to the design. It helps to generate and optimise
membership functions as well as the rule base from the
simple data provided. Combining the learning power of
neural network with knowledge representation of fuzzy
logic gives advantage to NeuroFuzzy system. This paper
presents the design and simulation of FLC using Fuzzy
Logic Toolbox in MATLAB [5]. The tuning of fuzzy
inference system is carried out by back propagation
algorithm based on collection of inputoutput data of BOF
process at Vishakhapatnam steel plant, Vishakhapatnam,
A.P, India. ANFIS has provided a new method for
solving the problem of prediction and control of end point
carbon content of complex BOF process.
II. BASIC OXYGEN STEELMAKING PROCESS
Basic oxygen furnace steel making process is one of
the key processes in steel industry. Basic oxygen furnace
is widely preferred and effective steel making method due
to its higher productivity and low production cost. The
basic oxygen furnace is a steel shell lined with refractory
materials. The body is slightly cylindrical, open at the top
to receive raw materials and the oxygen lance which is
%euro Fuzzy Modelling of Basic Oxygen Furnace and its comparison with
%eural %etwork and GR%% Models
M.V.V.N. Sriram
1
, N. K. Singh
1
, G.Rajaraman
2
1
Department of Mechanical Engg. & Mining Machinery Engg, Indian School of Mines. Dhanbad, India
2
Quality Assurance and Technology Development, Visakhapatnam Steel Plant, Visakhapatnam,India
(malladisriram@ismu.ac.in)
978-1-4244-5967-4/10/$26.00 ©2010 IEEE