Research On Rolling Force Prediction Method Of Six-High Reversible Cold Rolling Mill

The production technology of strip cold rolling is an important symbol to measure the development level of steel industry.The shape quality and thickness precision of the finished products are closely related to rolling force.The rolling process needs to go through three stages:acceleration,steady speed and deceleration.With the continuous rolling,roll wear will cause the change of roll state,if the roll is not replaced in time,the cold rolling process can not be carried out normally.In this paper,various methods for predicting rolling force have been studied,taking the single stand six high reversible cold rolling mill,with the aim of improving the precision of the rolling force.According to the speed variation characteristics,it can be simply divided into steady speed section and variable speed section The coefficient of friction and the deformation resistance directly determine the accuracy of the rolling force calculation,basing on the basic principle of cold rolling process.Analyzing influencing factors of the coefficient of friction and the deformation resistance is to establish the corresponding fitting formulas.The parametric model library is established according to different widths,alloy numbers of the strip.It is using the Bland-Ford-Hill formula that to obtain the predicted value of rolling force.In order to predict the rolling force in the whole rolling process,the neural network was introduced,and the data of the steady speed and variable speed sections were analyzed.The addition correction and compensation model is first studied,combining mathematical method and neural network to predict rolling force.The calculation results of the model library are taken as the main value of the predicted value of rolling force.At the same time,a single hidden layer neural network model is built up to predict the difference between principal values and measured values,which compensated the prediction deviation of rolling force.Meanwhile,by changing the network output parameters,the rolling force is regarded as the output of the network,and the rolling force is predicted with the same network structure.On the basis of previous researches,a GA-BP neural network rolling force prediction model is built,which making use of the GA algorithm to optimization the weights and thresholds of neural networks.Last but not least,aim at the correlation of parameter changes in the roll changing cycle,the long-short-term memory(LSTM)network is applied to the rolling force prediction in the whole rolling process,which takes full account of the time series information of sample points,closer to the actual cold rolling production process.According to the marked roll change information,52136 data points are divided into 50 data sets,the first 40 sets of data are used as the training set,and the last 10 sets are used as the prediction set data.Comprehensive comparative analysis is performed from the time step,the number of neurons in the hidden layer,the learning rate,etc.,to determine the appropriate parameter values.The results show that prediction accuracy of the LSTM network model is the highest,which can be reduced to about 3%.