基于遗传神经网络的GH4169高温合金摩擦焊超声检测缺陷识别  

The blemish identify in ultrasonic testing of friction welded joints made by GH4169 heat metal alloys based on GA-BP model

作　　者：尹欣[1] 王玉[1] 赵俊敏[1] 张辉[1] 

机构地区：[1]西北工业大学机电学院,西安710072

出　　处：《机械设计与制造》2008年第5期113-115,共3页Machinery Design & Manufacture

摘　　要：摩擦焊接过程中,由于焊接参数波动或一些非正常因素的介入,接头中容易产生未焊合及伪结合缺陷,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,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。There are welding defects, such as lack of bounding and weak bounding defects ,occurring during friction welding because of fluetuation of welding parameters or other abnormality reasons ,however the traditional BP network exists some problem,such as the slowly convergence speed and easily sinking into the partial extremum spot,during the period of carrying on the blemish identify,it will introduce the germ arithmetic into the BP network to turn a beginning to start the power value, putting forward the improvement BP network of the blemish identify, Pass to verify the blemish signal of the some GH4169 heat metal alloys detected by a super voice C ,which expresses that the method makes blemish qualitative analysis classification acquired higher accuracy,and it can lower artificial and subjective factor consumedly to the influence that the blemish identify.

关 键 词：缺陷识别 BP网络 遗传算法 定性分类 

分 类 号：TH16[机械工程—机械制造及自动化] TP183[自动化与计算机技术—控制理论与控制工程]