小样本字典学习的兰姆波模态识别方法*

针对兰姆波多模态识别问题,提出了基于小样本字典学习的模态识别方法。将多层复合板的频散特性看作一个线性时不变系统,首先,根据频散知识模拟各个模态传播特定距离后的信号,提取走时和能量特征创建字典；其次,获取待测信号的走时特征,通过查询字典来识别兰姆波模态；最后,根据能量参数估计结果,实现待测信号中各模态信号的分离和重构。通过对三层粘接的AAA板(铝板-亚克力板-铝板,每层厚度为2mm)中传播距离为0.3m、0.5m的直达波和反射波的实验验证,结果显示该方法对A0、S0模态的有效识别和各个波包信号的准确重建。

Lamb mode identification method based on small samples dictionary learning

In this paper, a Lamb mode identification method based on small samples dictionary learning is proposed. Firstly, according to the dispersion characteristics of multi-layer plate, Lamb waves of each mode propagating at specific distances are simulated, and the signal features (time of flight, energy distribution) are extracted to create a dictionary; secondly, the traveling time of the signal to be identified is extracted, then Lamb modes are identified by searching the dictionary; finally, the energy parameters are estimated, and signal separation and reconstruction is realized. Through the experimental verification of direct wave and reflected wave with propagation distance of 0.3m and 0.5m in AAA plate (aluminum-acrylic-aluminum plate, thickness of each layer is 2mm), A0 and S0 modes can be identified and reconstruction effectively.