Data Mining Forecasting Quality Of Hot Rolled Strip

Long-term production has accumulated abundant data which contains knowledge about hot-roll product's performance in steel enterprise. The data can be used to set up mechanical properties forecasting model in which chemical composition and rolling parameters as inputs, mechanical properties as outputs. we can use the model on off-line mechanical properties prediction and on-line mechanical properties control which is good for improving product quality and production efficiency.The thesis study on mechanical properties prediction of hot-roll product according to the actual situation of steel enterprise and KISC. Because of the complexity of data origin, we must extract, convert, integrate the needed data. Then we establish the data market on hot-roll product's performance by mature data warehouse and modeling. Three kinds of model based on multi-line regression and Artificial neural networks(ANN) are established by the data market on hot-roll product's performance:the model based on multi-line regression, the model based on BP neural network and the model based on RBF neural network. The prediction result indicates that 95% of the mechanical properties prediction error of BP model is in 5%, and 100% of the mechanical properties prediction error of BP model is in 10%.It means that BP model has good precision in mechanical property prediction and is suitable to extend.In the end, we investigate the influence between the chemical composition, rolling parameters and mechanical property by BP model, which provided the reference value for the on-line mechanical property prediction and the optimization of product rolling.