Research On Plate Profile Setting And Adaptive Method Of PC Rolling Mill Basedon Data Driven

With the upgrading of steel product structure and the improvement of product quality,new requirements are put forward for the calculation accuracy of rolling process mathematical model.However,the hot rolling plate profile models were mostly built based on a great deal of simplifications and hypotheses,and there were bottlenecks in calculation accuracy of these models.In recent years,with the improvement of computer hardware performance and the accumulation of rolling process data,the idea of building plate profile models based on data-driven has emerged as the times require.Successful model-setting rules are extracted from a large number of rolling process data by data mining technology,and mechanism models are replaced by data mining methods to avoid endless exploration of deep rules of plate shape.While perfecting the rolling model construction method,self-learning is also a good way to improve the accuracy of model setting.However,the traditional self-learning method of rolling model based on division of layer has the problems of model self-learning coefficients of adjacent layers are jumping greatly,discontinuous and other issues,the accuracy of self-learning coefficients needs to be improved.Therefore,the data mining methods including cluster analysis and case study were referenced.The hot rolling plate shape PC angle setting case library was established by historical data,and a new intelligent setting method of hot rolling plate shape PC angle was proposed.Firstly,some rolling cases with good strip shape were excavated through the cluster analysis on a large number of rolling historical data,and an initial rolling case library was built.Then,the iterative learning was applied in the case library during daily production,and the better cases were updated to the case library according to their layers.When using it,the rolling case closest to the current strip condition was selected from case library by similarity computation,and the corresponding PC angle was taken for the online plate shape setting.In addition,a new self-learning method for rolling model based on continuous surface is proposed.The layers are represented by the smooth surface which is characterized by feature points,based on the obtained model self-learning coefficients from some part specifications,the continuous surface is used to reconstruct an approximation of the unknown function,and the fast specification extension of model self-learning coefficients is realized.From discontinuous adjacent layers to continuous derivable in multi-dimensional space,the self-learning coefficient of the model can be accurately reached to any point in space,which has a qualitative breakthrough in improving the accuracy of model setting.The above research has been successfully applied to a large hot strip mill in China.On-line application shows that the accuracy of plate shape control and the prediction accuracy of deformation resistance and rolling force is improved remarkably,which meets the requirements of high precision shape control index and stable rolling production of hot strip.