Intelligent roll contour optimisation method based on SVR surrogate model and big data

Abstract

Roll contour design is a core technology for strip shape control; however, its application is hindered by challenges such as poor adaptability across multiple product specifications, difficulty in multi-objective optimisation, and low simulation efficiency. Existing studies typically rely on finite element simulations to optimise roll contours, but these simulations are time-consuming and struggle to cover the full range of specifications. Artificial intelligence and surrogate modelling techniques offer a novel approach to addressing this issue: by constructing machine learning based surrogate models of finite element simulations, computational efficiency can be significantly improved while maintaining accuracy. This article focused on a cold rolling mill, where optimisation weights were determined based on industrial big data. Multi-objective data were obtained through finite element simulations, and a surrogate model of roll system deformation was constructed using the support vector regression algorithm to achieve multi-objective roll contour optimisation. This approach is adaptable to a wide range of strip specifications and effectively mitigates shape defects, providing a feasible pathway for intelligent roll contour design.