Prediction of Steel Articles Mechanical Properties Through Raw Stock Chemical Composition and Smelting Parameters

The paper deals with the creation and application of a mathematical model designed for operational control of strength characteristics of items produced under conditions of open-hearth steelmaking. Variations in the chemical composition of the raw material as well as in the parameters and modes of smelting are used as controlling influences. The mathematical model is based on a neural network trained on a data set collected under batch production. Training, validation and testing of neural networks have shown that chemical composition parameters alone are insufficient for predicting the mechanical characteristics of manufactured products. The dependences of impact toughness of representative samples on chemical composition and melting conditions obtained as a result of virtual computer simulations allowed us to formulate recommendations to prevent product defects. It is noted that the dependences of strength characteristics of manufactured products on their chemical composition and melting conditions revealed through modeling have a complex non-linear character which sometimes does not coincide with the known results obtained on laboratory samples in isolation from the real conditions of serial production. It has been shown that the developed neural network model enables selecting the chemical composition and smelting parameters which eliminate reject and maximize the strength characteristics of the resulting products. It also helps to optimize the chemical composition of products and save the most expensive alloying elements.