?
Об управлении химическим составом сырьевого материала и режимом плавки для обеспечения требуемых механических свойств стальных изделий серийного производства
A The goal of the work is to create a mathematical model suitable for operational control of
the strength characteristics of the resulting steel product in the conditions of serial steelmaking.
Existing approaches based on the results of testing prototypes obtained in laboratory conditions
are not suitable for this purpose, since in the conditions of serial steelmaking, the strength
characteristics of products, in addition to their chemical composition, are affected by the structure
of the metal and many other melting conditions. Approaches in which the structure of the
metal is taken into account when making predictions also cannot be used, because obtaining
parameters of the metal structure is possible only after casting and solidification of the
steel, when operational control actions on the results of melting are no longer possible. The
main idea of the study is to train a neural network on those data from a serial production
process that directly or indirectly affect the mechanical characteristics of the resulting products
and, thus, take into account the structure of the metal in an implicit way. It is noted that
data collected under the conditions of existing mass production inevitably contain many
statistical outliers, so the datasets were thoroughly cleaned using the author’s algorithm,
which made it possible to create a neural network model suitable for practical use. Using the
developed neural network model using the freezing method, the dependences of impact
strength on the operating modes of the open-hearth furnace, melting conditions and chemical
composition were plotted in graphical form. The study of the neural network model made
it possible to identify some regularities of the simulated process, in particular, to establish
that in the conditions of open-hearth production, the chemical composition does not play a
primary role in the formation of the strength characteristics of products. As a result of studies
of the neural network model, recommendations were obtained for increasing the impact
strength of manufactured products and for removing some of them from reject by changing
the melting conditions and the chemical composition of the metal.