Presetting control of bending force in cold-strip rolling process based on the SSA-BP prediction model

Abstract

The flatness quality of cold-rolled strips is regarded as a crucial quality indicator within the production of steel strips. The accuracy of the bending forces applied to the work and intermediate rolls directly affects the control quality of the strip flatness. Traditional models are based on assumptions and parameter simplifications, which result in lower computational accuracy. To explore the relationship between various process parameters of the rolling process and their impact on the strip flatness, this article proposes a data-driven online setting model for bending force. The model integrates the isolation forest algorithm, sparrow search algorithm, and backpropagation neural network. To eliminate the disturbance caused by non-effective data on prediction results, the isolation forest algorithm is employed for outlier detection. Through these algorithms, the analysis, processing, and feature extraction of extensive measured rolling data are achieved, constructing a data model for the optimal setting. Finally, the validation process was conducted by employing acquired rolling datasets from a 1450-millimetre cold continuous rolling mill. The results demonstrate that the sparrow search algorithm and backpropagation model outperforms traditional roll deformation analysis methods and the standard backpropagation neural network model with respect to convergence speed, precision, and robustness.