基于EMD的胶合板损伤声发射信号特征提取及神经网络模式识别

摘要：针对胶合板损伤声发射信号的非平稳性和损伤类别特征相互重叠的实际情况，提出了基于经验模态分解(Empirical Mode Decomposition, EMD)和BP神经网络相结合的信号特征提取和识别方法。首先对损伤声发射信号进行EMD分解，筛选出包含主要信息的本征模态函数（Intrinsic Mode Function, IMF）分量；其次构建以各IMF分量的能量占比作为表征各损伤信号的特征向量；最后以提取的特征向量为输入样本，建立BP神经网络模式分类器对四类胶合板损伤信号进行识别。五层胶合板损伤的实测数据表明，该方法能够准确地提取出声发射信号特征并对其损伤类型进行有效地识别。

Feature extraction and pattern recognition of acoustic emission signals generated from plywood damage based on EMD and neural network

Abstract: Aiming at the non-stationary features of acoustic emission (AE) signals generated from plywood damage and overlapping of damage features in practice, a method of feature extraction and pattern recognition was proposed based on empirical mode decomposition (EMD) and BP neural network. Firstly, the original AE signals were decomposed by EMD, and intrinsic mode function (IMF) including the main feature information were selected. Secondly, energy ratios of IMF were constructed as feature vectors to identify damage signals types. Finally, BP neutral network pattern classifier was established to identify four types of plywood damage signals. The measured data of five-plywood damage show that the method can extract AE signals characteristics precisely and identify damage types efficiently.