Prediction Of Quality And Performance Of Hot Rolling Products Based On Deep Learning

Hot rolling technology is a complex industrial process in steel rolling production.In order to improve the quality and performance of hot-rolled products and to meet the requirements of the products produced,it is essential for the development of steel enterprises to establish an accurate and reliable prediction model for the quality performance of hot-rolled products.Since the production of hot rolled products requires multiple processes,each process contains factors that affect the quality of the product.In addition,the data collected from the factory is easily polluted by industrial noise.So how to establish a stable model is a difficult problem in the current research field.Therefore,this thesis focuses on how to build a more accurate prediction model.This thesis uses data mining technology to develop the data provided by the enterprise,and then converts it into useful knowledge information which is used in the process of predictive model establishment.The mechanical parameters of hot rolling,such as elongation at break,yield strength and tensile strength,are taken as the output variables of the model.And the influencing factors related to product performance are determined by correlation analysis.Data marts are then built through data preparation,data cleansing,and data integration.Finally,the data mining algorithm is selected to establish the quality performance prediction model of hot rolled products.The thesis focuses on the prediction model of quality performance of hot rolled products based on deep learning.Deep learning has good nonlinear autonomous learning ability,and its deep network structure has great advantages in extracting data features.Combining the above advantages of deep learning,this paper first improves the selection method of input variables,then uses the deep feedforward neural network to establish a prediction model of the quality performance of hot rolled products,and compares it with the prediction model established by the shallow neural network.The experimental results verify the feasibility of deep learning in the establishment of prediction model of the quality performance.At the same time,considering the problem of feature parameter redundancy caused by the full connection between neurons in the deep feedforward neural network,the convolutional neural network(CNN)can greatly reduce the feature parameters to be trained in the model due to its two functions of local connection and weight sharing.Therefore,the performance prediction model based on CNN is established.In order to strengthen the interaction between input variable factors,a method of transforming one-dimensional data into two dimensions is proposed.Finally,in analyzing the parameters of CNN,a new network structure is obtained by merging two CNN structures.The new network structure is further improved,and the quality performance prediction model for hot rolled products is established by using the improved structure.The simulation results show that the improved CNN structure increases the prediction accuracy and correlation of the model.