改进型BP网络在优化焊接工艺参数中的应用  

Application of Improved Neural Network in Optimizing Welding Parameter

作　　者：贾剑平[1] 徐坤刚[1] 李志刚[1] 

机构地区：[1]南昌大学机器人与焊接自动化重点实验室,江西南昌330031

出　　处：《热加工工艺》2008年第21期98-100,共3页Hot Working Technology

基　　金：国家863项目(2007AA04Z242);江西省科技攻关项目(2007BG09100)

摘　　要：运用改进的BP网络方法建立焊接工艺参数与焊缝成形的关系,提出了L-M(Levenberg-Marquardt)优化算法进行改进BP神经网络,采用六因素和两水平的正交试验法安排实验,保证了网络训练样本的代表性,构建了旋转电弧焊缝成形的预测模型。预测结果表明,该方法有较高的预测精度,最大相对误差不超过3%。该网络模型的建立为优化旋转电弧工艺参数提供理论依据。The optimal method of L-M （Levenberg-Marquardt） was proposed for improved BP neural network. The relationship between the process parameters and the formation of weld was estabilshed. The predictive model for the rotating arc was put forward, and orthogonal test were arranged using the six factors and two levels to ensure the representative sample of network training. The results indicate that the method has the high predictive precision, and the max relative error does not surpass 3%. Establishment of the network model provides a theoretical basis for optimizing process parameters of rotating arc.

关 键 词：BP网络 预测模型 工艺参数 应用 

分 类 号：TG444.3[金属学及工艺—焊接]