Study On Processing Parameters Optimization Of Three-rolls Planetary Rolling

With the enlargement of the application fields and increasing quality requirement of the copper tubes, higher demand are imposed on the processing of copper tubes.The continuous casting & planetary rolling method is an advanced technology for the manufacture of the copper tubes, with three-roll planetary rolling as the key process,which is a one-pass rolling process with large deformation. Although the processing parameters have a direct influence on product quality, it is very difficulty now to have an effective control of them during the planetary rolling process. Therefore, it should have important value in instructional significance and practical applications to make a study on processing parameters optimization of the three roll planetary rolling.A method combined of orthogonal design of experiments, neural network and genetic algorithms is adopted in this paper to optimize three-roll planetary rolling parameters, which overcomes the deficiency of traditional methods. Firstly, the structure of the three-roller planetary mill and its mechanical transmission principle are introduced. The rolling principle of every deformation interface and the deformation characteristic of tubes during the rolling process are then discussed.Taking three-rolls planetary rolling force as the optimization goal, offset angle,rotational speed of rollers and friction coefficient are confirmed as its primarily influencing factors. After that, using the finite element software--MSC.Marc, the finite element model of three-roll planetary rolling are built up, based on which a series of numerical simulation of the rolling process is carried out, according to the simulation scheme by orthogonal design of experiments. Based on simulation results, the rolling force under different processing conditions are calculated, which are then used as the training set of neural network.Neural network and genetic algorithms are the core techniques of the entire optimization. Using BP Neural network, the mapping relationship model between process parameters and rolling force is established. The effects of offset angle, rotational speed of rollers and friction coefficient on rolling force is predicted and analyzed with the established model. In order to improve the generalization and precision of the neural network model, the training set is preprocessed before training. The optimal number of neurons in hidden layer and training algorithms are determined. Genetic algorithms is adopted to search optimized processing parameters combination. To ensure the optimization of results ,the steepest descent method is used for iterative calculation, and the optimized processing parameters for the minimum rolling force are obtained. Finally, with programming technique of Visual Basic, MATLAB and SQL.Server, a optimization flat, called ED-NN-GA, has been built up. Based on the design theory of orthogonal table and COM technique, experimental design, neural network and genetic algorithms modules are completed. For better efficiency of the program, a database to storage training set and neural network mode is set up in this paper .The effectiveness of the optimization flat is validated by spot simulation and analytical calculations.