Predictive modeling of mechanical properties of welded joints based on generalized dynamic fuzzy RBF neural network

Generalized dynamic fuzzy neural network model was established to predict the mechanical properties of welded joints. Structure of the model is no longer in default modeling, but on a sampleby dynamically adaptive learning process. By introducing elliptic basis functions to expand the receive domain to function , increased fuzzy rules was based on the systematic error and fuzzy rules ε completeness, and the RBF unit width determination criterion was based on fuzzy rules ε completeness. The fuzzy rule of model pruning was based on their importance which was evaluated by error reduction rate. By using three different thicknesses and different process TC4 titanium alloy TIG welding test group, 17 sets and 5 sets of training and simulation sample data were obtained for modeling and simulation. The results showed that the model can accurate prediction on the mechanical properties of welded joints.