超声检测缺陷分类的小波分析与神经网络方法

根据金属超声检测中缺陷脉冲回波为非稳态信号的特点，提出了一种基于小波变换和模式识别技术的缺陷宁性分类方法，重点研究了利用小波变换提取反映缺陷性质的特征值以及动用模式识别技术对特征值进行缺陷定性识别的方法。为验证上述方法，设计了实验系统，同时对信号的采集异常信号的剔除等问题进行了研究。利用实际焊接试样进行了实验，经小波变换提取缺陷特征值，然后采用ＢＰ（ｂａｃｋ ｐｒｏｐａｇａｔｉｎｏ）神经网络，使缺

Application of Wavelet Analysis and Artificial Neural Network Pattern Recognition to Flaw Classification in Ultrasonic Testing

?According to the nonstationarity of pulse echo signals of flaw in ultrasonic testing, a method of flaw classification based on the com bination of wavelet transform with pattern recognition was presented . Method of extracting characteristic values reflecting the flaw properties usin g wavelet transform and the method of qualitatively recognizing the characterist ic values using pattern recognition were studied. An experimental system was used to test the method abov e, by which some real weld flaws were processed. Firstly the feature values of f laws were extracted with wavelet transform, then the flaws were classified with back propagation neural networks. The problems of signal data acquisition a n d the eliminating of pulse interference signals brought out from data acquisitio n were also considered carefully during the experiment. The results show that by this method human effects on qualitative recognition of flaws can be reduced to some extent, and high accuracy of flaw classification can be obtained.