以碳纤维复合材料的分层、孔隙和疏松缺陷的超声波检测信号为研究对象，对包含缺陷信息的复合材料超声波检测信号进行小波包变换，从近似系数及细节系数提取样本的特征值。建立并训练了一种用于实现缺陷识别的BP神经网络，该网络使用Levenberg-Marquardt算法可以快速地完成对数据的处理。使用该网络可进行缺陷类型的识别。

Based on signal of carbon fiber composites defect such as lamination, porosity, looseness in ultrasonic testing , this paper performs wavelet packet transform on ultrasonic testing signals for carbon fiber composites that contain defect information, extracts sample-features from approximation coefficients and detail coefficients. it builds and trains a BP neural network for defect identification. The network uses Levenberg-Marquardt algorithm to quickly process the data. It identifies the defect type by means of BP neural and achieves good effect.