Design Of PQF Pipe Mill Rolling Process Based On Genetic Neural Network

China as a large industrial state, iron and steel industry occupies the national economy pillar status. In recent years, China is in a stage of transition from the big industrial country to the industrial powerhouse. The development of national economy is rapid, and the international steel market demand continues to expand. This also poses a great challenge to the seamless steel pipe industry. PQF (Premium Quality Finishing) rolling tube machine is one of the most advanced process technology of rolling tube machine in the world. In August 2003, the world’s first set of PQF rolling units put into operation in Tianjin. PQF as the new breakthrough in seamless steel tube industry, its rolling process design is widespread concerned.PQF rolling process is a multi-factor, nonlinear and complex problem. At present, people both at home and abroad only use finite element method to do theoretical study of PQF process methods, and haven’t get the process method which adapts to the actual production needs. We have no experience to draw on. BP neural network has strong adaptive, self-organizing, associative memory, parallel processing, and other characteristics. It is suitable for dealing with complex issues, and is widely used in the field of nonlinear optimization, function approximation, forecasting assessment, and other areas. Therefore, this paper uses BP neural network to model, and uses data provided by MEER Company to train the network. Thus, it gets the PQF rolling process model, and forecasts the PQF automation control parameters. After several experiments, this paper determined the number of layers and parameters of the model. Because BP neural network has the slow convergence shortage and has the weakness of easy to fall into local minimum value, this paper uses the global search characteristic of genetic algorithm to optimization of BP neural network’s initial weights. It is in order to avoid local convergence of BP neural network, and improve the speed of convergence.According to the pass series provided by Tianjin Pipe Group Corporation, this paper builds a PQF rolling process model, and uses MATLAB software to do training, validation and prediction on this model. The experiments show that using the genetic neural network model, established by this paper, in PQF rolling process design can preset PQF automation control parameters and achieve better prediction.