Abstract: For high nonlinear, complex interaction of many factors in welding process, it is difficult to predict the mechanical properties of welded joints. At the same time to overcome the lack of common back propagation(BP) neural network. A fuzzy radial basis function(RBF) neural network model is built to predict mechanical properties of welded joints based on Fuzzy C-means cluster and Pseudo-inverse method. Take the TC4 titanium alloy Tungsten inert gas(TIG) arc welding process parameters including：Welding current, welding speed, argon gas flux for the input parameters, take the mechanical properties of welded joints including： Tensile strength, bend strength, extensibility, weld hardness and heat affected zone hardness for the output parameters. The 27 sets experiment data are used to train this model, other 9 sets are used to simulate this model. Simulation results show that the model’s structural is stability, training speed is fast, adaptability is strong, robustness is good, prediction accuracy is high, it can be used to predict the mechanical properties of welded joints. Through mathematical analysis, it can express the rule between welding process parameters and mechanical properties of joints use functional from, also can be used to optimize the welding parameters, provide the basis to adjust and control the quality of welded joints.

Key words: fuzzy C-means cluster;fuzzy radial basis function(RBF) neural network;prediction;welding;modeling