混合聚类RBF神经网络焊接接头力学性能预测

构建混合聚类算法,与伪逆法结合建立RBF神经网络模型预测焊接接头力学性能.以TC4钛合金TIG焊接试验为基础,将焊接参数作为模型输入,焊后接头力学性能作为模型输出.通过仿真,该模型预测平均相对误差范围为1.74%~6.69%,具有较高的预测精度、适应性和泛化能力,能够预测焊接接头力学性能.采用数学解析对所建模型分解,得到焊接工艺参数与接头力学性能之间映射关系的函数表达式,可优化焊接工艺参数.利用焊接专业知识对模型的径向基单元参数进行调整,提高了模型的预测精度,为将焊接专家知识融入RBF神经网络模型开辟了新方法与途径.

Prediction of mechanical properties of welding joints by hybrid cluster fuzzy RBF neural network

The build of hybrid clustering algorithmcombined with the pseudo-inverse method was carried out to establish the RBF neural network model to predict the mechanical properties of welded joints. Taking TC4 titanium alloy TIG welding experiments as basis, the welding parameters was set as model input and the mechanical properties of welded joints was set as output. Through simulation, the mean relative error of the predictions ranged from 1.74 to 6.69%, indicating that the model has higher prediction accuracy, adaptability and better generalization ability to predict the mechanical properties of welded joints. The model decomposed by using the mathematical analysis method, can obtain a functional expression between the welding parameters and mechanical properties of the joint process. The welding parameters can also be optimized simultaneously. The utilization of welding professional knowledge was applied to adjust the RBF unit parameter of model, allowing an increase of the prediction accuracy of the model. It has opened a new way to take the welding expert knowledge into the RBF neural network model.