Application of Machine Learning Methods for the Prediction of Roll Force and Torque during Plate Rolling of Micro-alloyed Steel

Abstract

Machine learning technique is extensively used to establish the relationship between non-linear data sets which cannot be described mathematically and thus an exact analytic model is either intractable or too time-consuming to develop. During hot rolling, the effect of process parameters that cannot be captured in mathematical models, such as roll dimensions and its wear, the inter-pass time between rolling passes, temperature variation has been incorporated using multivariate supervised machine learning technique for accurate prediction of roll force and torque during plate rolling of micro-alloyed steel. An ensemble method was used to combine various machine learning techniques and average them to develop one final predictive model. K-cross validation of the model was carried out to validate the results and ensure the model gets the correct pattern of data. Root mean square error of ensemble roll force model was compared with roll force calculation using Sims theory. It was found that the machine learning model can predict the roll force and torque accurately as it takes care of various non-linear process variables which cannot be accounted for mathematically. The R-value of the machine learning model was > 98%, whereas it was 92.2% for roll force calculation using Sims theory.