基于小波包变换的复合材料超声波检测信号特征提取

本文以碳纤维复合材料常见缺陷分层、孔隙、疏松的超声波检测缺陷信号为研究对象，对超声波检测信号进行小波包变换，提取包含信号绝大部分能量的近似系数波形特征及细节系数的统计量作为样本的特征值。应用BP神经网络分类器进行分类识别验证，取得较好的识别效果。该方法能以较小的特征维数表征原始信号特点。

Feature extraction from carbon fiber composites ultrasonic signals based on wavelet packet transform

Based on signal from carbon fiber composite defects such as lamination, hole, looseners in ultrasonic testing, this paper performs wavelet packet transform on the ultrasonic signals to extract statistics of approximation coefficients and detail coefficients that contain a great part of signal energy as sample-features. Then it identifies the defect type by means of the BP neural. The method is found to achieve good effect and can specify characteristics of the testing signals by lesser dimension.