基于神经网络的超声无损检测缺陷定征方法的研究

针对超声无损检测中缺陷分类难、分类结果可靠性差等问题，给出了一种以神经网络为基础的缺陷特征分类方法。利用Ｆｉｓｈｅｒ线性判别方法对表征缺陷特征的时域信号的波形参数进行了选择，并将这些参数作为神经网络的输入矢量对网络进行训练，用该网络对缺陷特征进行了识别，结果表明：神经网络的识别率远大于传统的贝叶斯分类方法。

Neural network characterization of flaws detected by ultrasonic approach

A methodology for the recognition of weld defects, detected by ultrasonic approach, has been developed within two stages. In the first stage, a selection of the shape parameters defining the pulse echo envelope reflected from a genertic flaws, and defined in the time domain, is performed by Fisher linear discriminant analysis. In the second stage, the classification is carried out by a three layered RBFN nerual network, where the input values are the parameters selected by Fisher analysis. With regard to the neural network learning process, 135 weld defects have been considered, and the effectiveness of this approach has been confirmed by discriminant result.