Research On Cold-roll Plate Shape Control Based On Deep Learning

With excellent electrical and thermal conductivity and corrosion resistance,copper alloy thin strip is widely used in new energy vehicles,semiconductor integrated circuits,electronic connectors,5G communications and other fields.With the rapid development of high-end equipment in these fields,the requirements for the shape accuracy of copper alloy thin strip in the downstream industry are increasingly stringent.In this paper,a 530 mm single-stand six-high UCM reversible cold rolling mill is taken as the research object.Based on the actual production data of the mill,finite element simulation,deep learning and intelligent algorithm are used to study the efficiency coefficient of shape control and the presetting of roll bending force.The main research contents and conclusions are as follows:(1)Based on the equipment and process parameters of a single stand six-high UCM cold mill in a factory,a three-dimensional elastoplastic numerical simulation model of copper alloy thin strip rolling process was established by using ABAQUS finite element software,and the shape regulation efficiency coefficients of main regulating mechanisms such as work roll bending,intermediate roll bending and roll tilt were obtained.The influence of process and equipment factors such as width,thickness,reduction rate,work roll and intermediate roll diameter on the shape regulation efficiency coefficient of each regulating mechanism was analyzed,and its variation rule was clarified.(2)In the field measured data based on the principal component analysis method for outlier removal,noise reduction and other preprocessing,using data-driven partial least squares(PLS)method to obtain the actual conditions of the shape of the regulation efficiency coefficient,and the regulation effect of the regulation efficiency coefficient and the field actual regulation effect were compared and analyzed.(3)The regulation efficiency coefficient obtained by finite element simulation was used as the sample library of neural network learning,and the BP model,GA-BP model and DNN-GA-BP model were respectively used for model training.By comparing the prediction effect of the models,it can be concluded that the DNN-GA-BP model is superior to the traditional BP model and the GA-BP model.(4)A convolutional neural network(CNN)model containing two convolutional layers and two pooling layers was used to establish the profile presetting program of single-stand six-high UCM thin strip mill.Rolling experiments were carried out based on the presetting results.The results show that the profile deviation and the mean square error of profile obtained by CNN model are lower than those obtained by field model.