(1.山东大学 材料液态结构及其遗传性教育部重点实验室，
济南 250061;
2.山东大学 连接技术研究所，济南 250061)
摘 要: 提出了建立熔池正面信息与背面熔宽的关系模型的新方法，由神经网络训练均方根误差确定模型的输入节点及隐层单元个数，并根据这种方法建立了正面熔池几何形状参数与背面熔宽的关系模型，非训练样本校验结果表明，该模型具有较高的精度。
关键字: 焊接熔透；视觉传感器；背面熔宽；神经网络
information in TIG welding
（1.MOE Key Lab for Liquid Structure and Heredity of Materials,
Shandong University,Jinan 250061,China;
2.Institute of Material Joining,Shandong University,
Jinan 250061,China）
Abstract:A new methods to establish a model describing the relationship of t he weld pool geometry and the back width of weld pool with sufficient accuracy has been proposed. The unit-number of hidden layer and input layer ca n be decided according to the training error. The relationship of the weld pool geometry parameters and the back width of weld pool is established using this method. The test results show that the model has sufficient accuracy.
Key words: weld penetration; visual sensor; back width of weld pool; neural network
