The scheme of intelligent control system of cap-bending has been advanced in this paper using the neural network technology, based on the prominent problem that bending springback difficult to control accurately durin...The scheme of intelligent control system of cap-bending has been advanced in this paper using the neural network technology, based on the prominent problem that bending springback difficult to control accurately during the forming process of cap-bending. The key technology of real-time identification for material performance parameter and friction coefficient was researched, and the back-propagation neural network of real-time identification for material performance parameters and friction coefficient was established, which can real-time identify the needed material performance parameters through the real-time monitoring variable. Factors that affecting recognition results of neural network model were analyzed, such as influences of the selection of the sample date and the algorithm for identification result. Factors affecting neural network generalization ability were discussed, such as influences of the selection of the sample date and the node number of the hidden layer for generalization ability. The results provide a guarantee for improving the convergence accuracy and the generalization ability of network, and provide a basis for the building of intelligent bending control of network model.展开更多
文摘The scheme of intelligent control system of cap-bending has been advanced in this paper using the neural network technology, based on the prominent problem that bending springback difficult to control accurately during the forming process of cap-bending. The key technology of real-time identification for material performance parameter and friction coefficient was researched, and the back-propagation neural network of real-time identification for material performance parameters and friction coefficient was established, which can real-time identify the needed material performance parameters through the real-time monitoring variable. Factors that affecting recognition results of neural network model were analyzed, such as influences of the selection of the sample date and the algorithm for identification result. Factors affecting neural network generalization ability were discussed, such as influences of the selection of the sample date and the node number of the hidden layer for generalization ability. The results provide a guarantee for improving the convergence accuracy and the generalization ability of network, and provide a basis for the building of intelligent bending control of network model.
