Prediction of rolling force and spread in hot rolling process by artificial neural network and multiple linear regression

The aim of this study is to compare the performance of multiple linear regression (MLR) and artificial neural network (ANN) models in predicting rolling force and spread during free rolling in the hot rolling process. Accurate prediction of rolling force and spread in hot rolling is critical for ensuring homogeneous load distribution across rolling stands, enhancing energy efficiency, reducing failure stops, and achieving dimensional accuracy and high-quality final products. The data used in this study were generated through FEM analysis, with a portion of the results verified experimentally. The dataset includes variables such as material temperature, rolled material dimensions, reduction amount, and rolling speed, all of which influence rolling force and spread. A maximum acceptable error rate of 2.9% for spread and 6.7% for rolling force was determined. Both MLR and ANN models were applied to the dataset, and their prediction performances were compared using the mean square error (MSE). For rolling force estimation, the ANN model achieved a training R value of 0.9888 and a test R value of 0.9844, while the MLR model obtained an R2 value of 0.9651 and an adjusted R2 value of 0.9829. In spread estimation, the ANN model achieved a training R value of 0.9947 and a test R value of 0.9844, compared to the MLR model's R2 value of 0.9871 and adjusted R2 value of 0.9863. The results indicate that both models perform comparably well in estimating rolling force and spread. However, the artificial neural network model demonstrates a slight advantage, offering marginally superior prediction performance.