基于自适应Q-shift双树复小波分析的碳纤维复合材料缺陷识别

Ultrasonic nondestructive testing has been widely applied to identify flaws for carbon fiber reinforced plastics (CFRP) so as to guarantee its trouble-free service. The acquired ultrasonic echo signals are often surrounded by complex noise due to the special physical properties of CFRP. Moreover, the flaw signal features are not obvious. Therefore, traditional wavelet based methods are failed to suppress noise and extract features for the ultrasonic echo signals of CFRP. To solve the problems, this project tries to utilize dual tree complex wavelet transform (DTCWT) to analyze the ultrasonic flaw signals of CFRP. The main research content includes: (1) Design adaptive Q-shift filters meeting no aliasing, perfect reconstruction and orthogonality, and after that propose the adaptive Q-shift based DTCWT method. (2) Propose the DTCWT based block-thresholding denoising method with Stein's unbiased risk estimate to eliminate the mixed noise of original ultrasonic signals. (3) Propose the DTCWPT based local frequency energy feature extraction method to accurately characterize different flaw features. Finally, we train a classification model based on the proposed methods to implement flaw identification. The expected research outcome would provide a new approach to automatically and correctly recognize different kinds of flaws for CFRP.

为保证碳纤维复合材料的安全使用，超声无损检测被广泛用于对其各类缺陷进行识别。由于碳纤维复合材料特殊的物理性质，在检测时获取的超声回波信号往往噪声大、构成复杂，某些缺陷特征不明显，使用传统小波方法对这类信号进行降噪和特征提取时效果并不理想，进而导致了缺陷识别率不高。对此，本项目尝试使用双树复小波对碳纤维复合材料各类缺陷超声信号进行多尺度分析并解决上述问题。研究内容包括：①设计同时满足抗混淆、完全重构和正交化的自适应Q-shift滤波器，在此基础上提出基于自适应Q-shift滤波的双树复小波方法；②提出基于双树复小波的Stein无偏似然估计块阈值降噪方法，消除原始超声信号中的混合噪声；③提出基于双树复小波包变换的频带局部能量特征提取方法，实现各类缺陷超声信号特征的精确提取。最终通过构造和训练分类模型，利用所提出方法实现缺陷识别。预期成果将为实现碳纤维复合材料缺陷的自动和准确识别提供新的途径。

超声无损检测被广泛用于对碳纤维复合材料各类缺陷进行识别，然而获取的原始超声信号往往噪声大、构成复杂且缺陷特征不明显，使用传统方法对这类信号进行降噪和特征提取时效果并不理想，进而导致了缺陷识别率不高。对此，本项目使用了双树复小波对碳纤维复合材料各类超声缺陷信号进行多尺度分析、降噪和特征提取并完成缺陷识别，解决了传统方法的不足。主要研究内容包括：①设计同时满足抗混淆、完全重构和正交化的自适应Q-shift滤波器，在此基础上提出了基于自适应Q-shift滤波的双树复小波方法，对超声缺陷信号进行阈值降噪后提高了信噪比；②提出了基于双树复小波的无偏似然估计块阈值降噪方法和双变量收缩降噪方法，有效消除原始超声信号噪声的同时保留了信号细节结构与特征；③提出了基于双树复小波变换的频带局部能量特征提取方法和经验模态分解的特征提取方法，对分层、疏松和孔隙缺陷的平均识别率分别达到93.75%和91.22%，并降低了分类模型的训练时间。此外，项目组还对稀疏表示、灰度共生矩阵以及局部二值模式等方法和理论进行了拓展研究，后续将进一步考虑如何将相关技术有效应用于碳纤维复合材料的缺陷识别。总之，项目组较好地完成了研究目标，取得的成果具有较高理论和实际指导意义。