Attentional Residual Network for Necking Predictions in Hot Strip Mills

Abstract

In hot strip mills, prediction of the necking width of the hot strip is a fundamental step in the hot strip mill process. However, owing to the large number and complexity of the variables involved, this prediction remains a challenging problem. In this article, we propose a deep neural model with an attentional residual network that combines an attentional network to calculate feature importance and a residual network to estimate the necking value. When a hot strip mill dataset from a South Korean steelmaking company was evaluated, the proposed model showed a higher performance than several machine-learning methods. Furthermore, the importance of the features selected by the attentional network outperformed those by other feature selection methods. Our approach is useful for necking predictions and can be applied to determine feature importance.