运用BP神经网络对船用钢焊接收缩量建模研究

重新建立了实用的船用钢焊接收缩量计算模型．首先分析焊接变形产生的原因，然后采集实船焊接变形数据为样本，运用BP神经网络，分别建立了输入层单元数为2、隐层单元数为3、输出层单元数为1的平板对接焊及T形焊焊接收缩量的神经网络模型．介绍了网络的结构设计及其训练过程，算例及应用证明，本模型对实施船舶精度控制具有指导意义。

Study of mathematical models construction method for shrinkage of welded steel by BP neural network

It is important for ship production to form logical models for welding distortion. The cause of welding distortion has been analyzed. Based on the BP neural network, two mathematical models of transverse shrinkage due to butt weld and T-joint fillet weld, with input layer nodes 2, concealed layer nodes 3 and output layer node 1 are established, by using experimental data of practical ship fabrication. The network structure and the training process are explained, and examples show that the model is suitable for practice.