Predicting endpoint parameters of electric arc furnace-based steelmaking using artificial neural network

In steel manufacturing, the chemical composition of the raw material serves as the foundation for the properties of the final product. The objective of this study is to establish a prediction algorithm for estimating the highly nonlinear characteristics of chemical condensation of elements in an electric arc furnace. A multilayer feedforward neural network is used to estimate the fluctuations in parameters of molten steel. In this study, the prediction models utilize a synthetic dataset generated based on industrial data. An experiment was designed with seven multi-layer feed-forward neural networks with distinct architectures and optimization functions, including stochastic gradient descent and adaptive moment estimation, to evaluate the optimal architecture. The results demonstrated that the proposed method, which employs a mean squared error (MSE) loss function with a value less than 0.036, can effectively predict the amount of carbon, iron oxide composition, and temperature of molten steel, which are crucial quality parameters. This study proposes a novel method for optimizing steelmaking operations via the electric arc furnace route.