Skip to main content

Application of Neural Network Based Models for Optmization of the Rolling Process

Objective


Great amount knowledge of the use of ANN's as well as other data handling methods is gained. Many such tools or methods are developed. Several models concerning rolling of the steel strips are created or developed. Data availability and quality seemed to be one of the most critical factors; delays and deficiencies caused problems, especially in surface quality task.
The steel product manufacturers are under a growing
pressure to improve the yield and quality of their
products. To meet these challenges, highly advanced
rolling systems are being applied to new and updated
mills.

However, there are many process steps where exact
mathematical models do not exist and the process is
affected by human behaviour of the operator. Mill
presettings, for instance, are traditionally calculated
using mathematical models. However, continuous
recalculation of model parameter is necessary to adapt
them to actual process events. The application of neural
network based method is proposed to overcome these
shortcomings. Due to their ability to learn, neural
networks can find the structure within measured data. The
networks provide an effective way of building adaptive
and intelligent systems for industrial applications. They
are used in combination with traditional mathematical
models.

In process monitoring, the use of a Self Organising Map
(SOM) algorithm is a potential research approach. This
neural network method is based on unsupervised learning.
It creates a non linear mapping from a multi dimensional
input space to a usually two dimensional surface. SOM is
especially suitable in analysing and modelling complex
processes, with many process parameters as well as input
and output parameters, such as strip and plate rolling
processes.

Experiments and long term on line applications with
different types of neural networks indicate a significant
improvement in the pre calculation of the roll forces in
the finishing mill. In the present project, neural
network computation methods will be applied to improve
and optimise other important parameters such as width,
thickness, flatness and surface quality of rolled plate
and strip to achieve the superior quality of the best
suppliers in the world. The results will be recorded and
evaluated in pipe production, that is a major enduser of
coils. One third of the work is devoted to theoretical
considerations of relationships and dependencies within
and between processes and the tools needed for modelling.

The models to be developed in this project will consist
of new analytical and hybrid ones describing the rolling
process and the material properties. Additional expert
knowledge of the rolling process and knowledge of the
actual process state and its stability will be gained.
More experience in on line adaptation of neural networks
in the field of process control will be attained.

The consortium comprises two endusers of systems, one
enduser of coils, one system developer and supplier, one
developer and two R&D performers.

Funding Scheme

CSC - Cost-sharing contracts

Coordinator

Rautaruukki Oy
Address
155,Rautaruukintie
92101 Raahe
Finland

Participants (6)

Helsinki University of Technology
Finland
Address
2 C,rakentajanaukio
02150 Espoo
Krupp Hoesch Stahl AG
Germany
Address
12,Eberhardstraße
44145 Dortmund
MEFOS - THE FOUNDATION FOR METALLURGICAL RESEARCH
Sweden
Address
Metallvaegen 2
971 25 Lulea
Siemens AG
Germany
Address
60,Schuhstraße
91052 Erlangen
TECHNICAL RESEARCH CENTRE OF FINLAND
Finland
Address
3,Kemistintie 3
02044 Espoo
Technische Universität Freiberg
Germany
Address
2,Berhard-von-cotta-straße
09596 Freiberg