| In recent years,in the environment of scientific and technological innovation and development,the steel industry has gradually shown the characteristics of intelligence.Due to the rapid development of China’s steel industry,the quality of steel products is improving day by day,and customers’ demand for steel products is gradually diversified and personalized.Accelerating the intelligent production of products and improving market competitiveness and production efficiency have become urgent needs for the development of steel enterprises.In order to realize the high-quality and rapid development of China’s steel manufacturing industry,it is necessary to realize the transformation from manufacturing to "intelligent manufacturing".As the primary process in the forming process of seamless steel pipe,in traditional production,it is necessary to calculate the process parameters of the piercer as the set parameters according to the requirements of product specifications,and then adjust them through trial rolling with the help of production experience.The adjustment decision of key process parameters depends on the knowledge reserve and cognitive level of workers,and the debugging period is long,and trial production is needed during the debugging process.In order to solve the problem that the process parameters of the rolling mill have a great influence on the quality of the seamless steel tube in the two-roll cross-rolling piercing production process,and the accuracy of the set values calculated by the traditional mechanism formula is not high,this paper takes the process parameters of the two-roll cross-rolling piercing equipment as the research object,establishes the process parameter prediction model of the cross-rolling piercing machine based on deep learning,and makes a prediction study on the process parameters of the cross-rolling piercing machine.The main contents of this paper are as follows:(1)Firstly,the production process of seamless steel pipe piercing and the piercing deformation process are introduced.According to the theory of skew rolling,the main technological parameters affecting the ellipticity,uneven wall thickness and outer diameter of capillary in rolling mill are: roller spacing,guide plate spacing and plug extension.The above three parameters are set as the output variables of the model,and then six input variables are determined according to the traditional calculation formula of the output variables and combined with the actual production experience,namely: tube blank diameter,capillary diameter,plug diameter,feed angle,wall thickness and roll exit cone angle.(2)A process parameter prediction model of cross rolling punch based on pigeon group improved RBF neural network was proposed.The process parameters(roll spacing,guide spacing and tip elongation)of the two-high diagonal rolling punch were predicted by using RBF neural network.The center,variance(width)and the connection weight between hidden layer and output layer were optimized by pigeon swarm algorithm.The model was trained and verified by using production data.The simulation results show that the proposed model has high prediction accuracy and strong applicability for roll spacing,guide spacing and head elongation.(3)A prediction model of process parameters of cross rolling punch based on deep neural network is proposed.The deep neural network was used to establish the process parameter prediction model of two-high diagonal rolling piercer.The low batch gradient descent method and Adam algorithm were combined to make gradient estimation and correction when training the deep neural network,so as to optimize the training speed.The model was trained and verified with production data.The simulation results show that the prediction model based on deep neural network is superior to the traditional mathematical model and the improved RBF neural network based on pigeon group,and has higher prediction accuracy.(4)Design and develop an intelligent prediction interface for process parameters of crossrolling piercing.A visualised process parameter prediction calculation system was developed using MATLAB with practical industrial requirements as the design objective and embedding a core algorithm of multi-input and multi-output deep neural networks. |
