Study On Mathematical Model Of Rolling Force For Tandem Cold Mill Based On Neural Network

As expanding of specifications species and improving the accuracy of cold-rolled products on the market, it is quite necessary to master and optimize the rolling model. Rolling force is the most important equipment parameters and process parameters of set data in the cold tandem rolling. It is significant to study the practical and highly precise rolling force setup model.Traditional calculation on rolling force is based on mathematical model. Rolling force mathematical model has low precision because of the complication and variation of influence factors in rolling process. As a result, it can not provide higher precise rolling force values to tandem cold mill. This thesis take Baosteel cold tandem mill rolling force model as the research object. Focus on the adaptive correction of rolling force setup model and the model deviation compensation of rolling force based on neural network, which to improve the accuracy of setup model.The thesis analyzed the equation of Bland-Ford-Hill rolling force model and each sub-model in cold tandem rolling process and corrected the rolling force model using the exponential smoothing method of adaptive theory, mainly researched on the method of combination of the resistance to deformation and coefficient of friction adaption indirectly correct rolling force and the rolling force direct adaption. The artificial neural network combined with the adaption of rolling force method for model deviation compensation after the model adaptive revised. Established the8-17-1topological structure of the BP neural network as rolling force deviation model. According to the established BP neural network has the shortcomings of the slow convergence speed and easy falling into the local minimum, this thesis used the ant colony algorithm to initial weights of the neural network.Finally, this study have conducted the actual data simulation analysis using Matlab language for all the methods improving the precision of rolling force set model in this thesis. The simulation results have verified the feasibility and validity of these methods. Ant colony algorithm and BP neural network rolling force model has made the set error reduced to within±3%and improved the precision of rolling force set model significantly.