The Research Of Hot Rolling Sharp Setup Model Based On BP Neural Network

In recent years, China's steel production capacity of a serious surplus, steel price index repeatedly hits the bottom, the whole steel industry is in deeply loss and companies have to face survive crisis. The face of such difficulties, iron and steel enterprises in order to survive in such a tragic market environment, and only continue to improve their product quality, quality to seize the market. The key indicators for evaluating the quality of hot-rolled products include the accuracy of the plate shape, and the accuracy of the plate shape has become one of the focuses of the study by various steel producers.In this paper, our research is based on ANSTEEL 2150 mm hot rolling product line and LVC rolling mill. We adopt many methods to find defects of current sharp setup model, which include analyzing the theory and technology of strip steel sharp control, analyzing the theory of LVC rolling mill, studying the current sharp setup model and summarizing product data of actual product process. To improve the sharp accuracy, we import the BP neural network technology into the study and develop a new model based on BP network.Based on full study on current sharp setup model, find key factors affect sharp accuracy, which used as the input node of BP network model, and the optimal network structure of BP neural network is determined experimentally. We complete the BP neural network model program development, design the HMI screen related to BP neural network model to provide interfaces to data display, operator intervention and alarm, and develop the communication program. introduce the layers concept of mathematical model, using actual data for a single layer BP neural network training tests, and provide solutions of problems such as possible minimum, long training time and other issues. Finally, the accuracy of BP neural network model is evaluated by actual production data.The BP neural network model achieves very good results after applying in actual product line. We has trained and tested about more than 60% of the regular rolling specifications online, the standard deviation of bending force and shifting amout are lower than of the original setup model.