论文部分内容阅读
将神经网络技术用于CO2堆焊焊缝几何形状预测的研究中,在构造神经网络的过程中,以逼近焊缝上下表面轮廓的抛物线方程系数为输出变量,较为全面地反映了焊缝截面形态;对具有3个输入变量和4个输出变量的神经网络的结构和参数设计进行研究,得出了一组用于焊缝几何形状预测的最佳的神经网络参数;改进了基于Matlab神经网络工具箱的训练方法,有效缩短了神经网络的训练过程。
In the process of constructing neural network, the parabolic equation coefficient which approximates to the upper and lower surface contour of welding seam is taken as the output variable, which comprehensively reflects the cross-section morphology of weld The structure and parameter design of a neural network with three input variables and four output variables are studied. A set of optimal neural network parameters for the prediction of the weld geometry are obtained. The neural network based on Matlab is improved Box training methods, effectively shortening the neural network training process.