广义动态模糊神经网络焊接接头力学性能预测

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

建立广义动态模糊神经网络模型,用来预测焊接接头力学性能.模型结构不再是建模时预设,而是在对逐个样本的学习过程中动态自适应调整.引入椭圆基函数扩大函数的接收域,利用系统误差和模糊规则ε完备性作为模糊规则增加的依据,并将模糊规则ε完备性作为径向基单元的宽度确定准则.以误差减少率评价模糊规则的重要性,并以此为依据对模型的模糊规则进行修剪.采用三种不同厚度、不同工艺TC4钛合金TIG焊接试验,获得17组训练样本和5组仿真样本数据,建模并仿真.结果表明,该模型能够对焊接接头力学性能进行较为准确的预测.

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.