Edge drop prediction of hot-rolled silicon steel based on a mechanism and data model

Abstract

In this article, a high-precision edge drop prediction model for hot-rolled silicon steel based on a rolling mechanism and deep neural network (DNN) is proposed. Considering the initial roll shape, roll wear and roll thermal expansion, a mechanism model of the roll system is established according to the hot rolling 4-high finishing mill. In order to reduce the data dimension and effectively improve the prediction accuracy, the random forest algorithm is employed to analyse the feature parameters that affect the edge drop of hot-rolled silicon steel. Finally, the DNN is introduced to establish an accurate silicon steel edge drop prediction model based on mechanism data and rolling data. The experimental results demonstrated that the proposed DNN model has strong generalisation and reliability with a squared correlation coefficient (R²) of 0.9656, and it is suitable for application in hot-rolled silicon steel lines for predicting edge drop. Furthermore, the proposed model has a certain reference value for the hot rolling process.