摘要
在焊接缺陷的超声检测中,对缺陷进行定性分析是超声无损检测与评价的关键内容,也是超声检测研究领域的热点和难点。针对焊接缺陷超声回波信号的特点,利用小波包变换提取反映缺陷性质的特征值,运用概率神经网络对缺陷进行识别,并与BP网络、RBF网络的识别结果进行比较。实际焊接缺陷的实验结果表明,概率神经网络的识别正确率高,训练和测试速度快,可靠性高。
In ultrasonic testing of welding flaws,the qualitative analysis for flaws is a key content,hot and difficult point of ultrasonic NDT and NDE.Aimed at the characteristics of ultrasonic echo-signals of weld flaws,wavelet packet transform was applied to extracting flaw features,and flaws were recognised by using probabilistic neural network （PNN）.Then,its recognition result was compared with that of BP and RBF neural network.The experimental results for actual welding flaws show that the recognition accuracy of PNN is high,its training and testing speed is fast,moreover,the reliability is also high.
出处
《仪表技术与传感器》
CSCD
北大核心
2010年第8期89-92,共4页
Instrument Technique and Sensor
基金
西安科技大学培育基金(200740)
