Modeling and validation of bending force for 6-high tandem cold rolling mill based on machine learning models

When producing high-end grades of cold-rolled strips such as precision thin strips and high-strength automobile steel plates, it is difficult to control the flatness due to their small thicknesses or high strengths, and it is easy to produce high-order flatness defects. To alleviate the strip's flatness defects, we propose a modeling method for optimal bending force presetting that uses the principal components analysis algorithm (PCA) and beetle antennae search algorithm (BAS) optimized Elman neural network (PCA-BAS-ENN) to improve the bending force presetting accuracy of each stand, which then improves the shape quality of the cold-rolled strip. We collected historical production data of a cold rolling mill to build a preset model. An experimental comparison with the GA-BP and PSO-BP neural network algorithms, commonly used in the rolling field, was conducted under the same data conditions. According to the experimental results, the PCA-BAS-ENN model has the highest prediction accuracy, and the prediction accuracy and performance outperform the GA-BP and PSO-BP neural network models. Industrial field applications demonstrate that the PCA-BAS-ENN model effectively alleviates the high-order and local flatness defects and has an essential role in reducing the strip's 2-order and 4-order flatness deviations and improving the flatness control accuracy.