BP神经网络在摩擦焊超声检测缺陷识别中的应用  

The Blemish Identify in Ultrasonic Testing of Friction Welded Joints Based on GA-BP Model

作　　者：尹欣[1] 刘爱敏[1] 樊成芳[2] 

机构地区：[1]郑州航空工业管理学院机电工程系,河南郑州450015 [2]西北工业大学机电学院,陕西西安710072

出　　处：《新技术新工艺》2007年第9期57-59,共3页New Technology & New Process

摘　　要：在摩擦焊接过程中,当材料焊接性差、焊接参数不当或表面清理不好时,在摩擦焊连接界面上也会出现一些"非理想结合"的缺陷。而传统BP网络在进行缺陷识别过程中存在收敛速度慢和容易陷入局部极值点的问题,本文将遗传算法引入BP网络优化初始权值,提出了改进后的缺陷识别BP网络。通过验证超声C扫描成像法检测到的某GH4169高温合金的缺陷信号,表明该方法使缺陷的定性分类获得了更高的准确率,可大大降低人为主观因素对缺陷识别的影响。During friction weld,there are 'nonideal combination' because of bad weldable material,weld parameters or surface cleaning. The blemish identify by traditional BP network has slow rapidity of convergence and local limit point. These defects are improved by the heredity calculate way. The results show that this method has high accurate rate for qualitative analysis classification of defects and reduces the effects of artificial identity on the defects.

关 键 词：缺陷识别 BP网络 遗传算法 定性分类 

分 类 号：TG456.9[金属学及工艺—焊接]