Modeling of ladle metallurgical treatment using neural networks

Abstract EN

In the steel industry, the refining process achieves the final chemical composition and temperature of liquid steel by adjusting the optimal quantity of additives and energy.

Generally a conventional charge calculation based on mathematical and thermodynamic models that provides considerable help is used, but it is difficult to model the highly complex nature of the interaction between process variables such as thermal losses and the dynamics of non-linear chemical reactions.

Neural networks are able to identify internal relationships through training examples.

In this work, an application of identification models using a linear approach and neural networks to predict the final chemical composition and temperature for the refining process is considered.

Using an industrial process data base, the dynamics of complex reactions are modeled using the back propagation learning algorithm.

This model is used as a charge calculation to predict the final process parameters.

The performance of the model is evaluated from new inputs and outputs.

Production and quality cost management is reduced by an optimal control of the input variables such as the weights of additives (FeMn , FeSi, and coke) and heating temperature (T).