Data-driven Flatness Prediction And Optimization Control For Tandem Cold Rolling

In this paper,the author conducts a research on the way to obtain the flatness actuator efficiency factors and establish the flatness prediction model of UCM cold rolling mill.In this study,a 1450mm five-stand cold rolling production line is studied as a research object,and the data-driven method is used as a mathematical tool.A calculation method of the flatness actuator efficiency factors based on OSC-PLS algorithm is proposed,and the flatness prediction model is established by KPLS-ANN algorithm.Meanwhile,based on the KPLS-ANN flatness prediction model,the gradient descent method is used to optimize the flatness preset.The main research contents of this paper are as follows:(1)According to principal component analysis and kernel principal component analysis theory,the abnormal conditions are analyzed,and the abnormal points of the cold rolling production data are deleted.(2)Combining the orthogonal signal correction and the partial least squares,according to the cold rolling production data,the influence law of the flatness actuators is analyzed,and the flatness actuator efficiency factors are obtained by the flatness variation and the actuator adjustment.The data-driven method and the traditional experiential method are compared to these application effect,and the accuracy of the efficiency factors is tested.The best efficiency factor is obtained.(3)Considering the influence factors of flatness,a kernel partial least squares method and its improved method are proposed,and the shape prediction models based on KPLS and KPLS-ANN algorithm are established.For the actual production data,the KPLS-ANN prediction model achieves high-precision prediction of the flatness.The KPLS-ANN prediction model shows a good predictive ability,with RMSE of 0.51IU,MAE of 0.34IU and MAPE of 0.09.(4)Based on the KPLS-ANN flatness prediction model,the the gradient descent method is used to optimize the flatness preset.The flatness preset is optimized by comprehensive adjustment of the work roll bending force,the intermediate roll bending force and the roll tilt.The optimization process effectively modified the initial setting parameters of the actuators and reduced the flatness.The flatness preset achieved good results.For the five rolling mills,KPLS-ANN has a good optimization ability which gets the flatness standard deviation of 2.22IU.Comparing with the initial standard deviation of 4.10IU,the optimization is remarkable.The research results of this paper have strong practicability for the flatness prediction and control process.The research results have certain guiding sense to the development of the flatness prediction model and the flatness control system.