Application of machine learning to predict and diagnose for hot-rolled strip crown

Based on the complex characteristics of nonlinearity, strong coupling, and multi-disturbance, it is challenging to develop accurate mathematical models for strip rolling systems which are limited by the degree of fitting between mathematical models and real models. A new class of solutions based on machine learning (ML) models was proposed to predict and diagnose the target variables for hot rolling, and their feasibility was fully demonstrated by analysing experimental data. In addition, the particle swarm optimization (PSO) algorithm was employed to optimize the proposed models to enhance generalization performance. By comparison with artificial neural networks (ANNs) and regression trees (RTs), results show that the support vector machine (SVM) model achieves the best prediction performance, with a root-mean-squared error (RMSE) and a correlation coefficient (R) of 1.5725 and 0.9809, respectively. Also, diagnostic results clearly demonstrate that 97.83% of prediction data have an absolute error of less than 4.0 μm. Therefore, the ML method based on data driven can be considered an effective solution to manage complex engineering problems. In addition, simulation results can achieve real accuracy requirements well and have important reference significance and value for practical applications when improving the quality of shape control.