Abstract: The control system modeling of penetration is studied in titanium-alloy electron beam welding. On the basis of analysing the characteristics of electron beam welding, the orthogonal test of three factors and five levels is designed, the test is done to obtain the values of molten width and penetration under different welding parameters, the values of molten width and penetration are used as the training sample to train the neural network, thus a BP neural network model between molten width and penetration is established, the molten width is input, the penetration is output. The model consists of a S-function hidden layer plus a linear output layer. Because the mathematical model for penetration is difficult to obtain, so the fuzzy controller is designed, its input variables are bias and change rate of bias, its output variables are the change of welding current, The controller has nine fuzzy-control rules. The BP neural network model and fuzzy controller are combined to establish fuzzy neural network control-system model of penetration of titanium-alloy electron beam welding, and unit step signal is used to carry out the simulation experiment of the model, the results show that the dynamic performance and steady-state performance of designed control system are eminent.

Key words: Electron beam welding, Fuzzy control, Neural network