基于主成分分析的纤维拉伸断裂声发射信号识别

摘要

For problems of acoustic emission signal of fiber tensile fracture such as nonstationarity and high overlap between signal characteristics, a model was presented for the feature extraction of acoustic emission signal and fiber type diagnosis. The model can be used to identify the type of fibers stretched. Firstly, different kinds of tensile fracture acoustic emission signals were preprocessed and decomposed by wavelet transform and ensemble empirical mode decomposition ( EEMD ) . Then, the frequency characteristics were extracted by the principal component analysis ( PCA) . Finally, least squares support vector machine ( LSSVM) was used to classify the characteristic frequency of the fiber stretched. Results show that the principal component analysis method can further select the eigenvector sets of the two kinds of fiber tensile fracture acoustic emission signals, and make the signal characteristics from high dimensional to low dimensional. At the same time, the correlation between the features is reduced, and the accuracy of recognition of fiber tensile fracture of AE signal is improved. The EEMD-PCA-LSSVM model has a total recognition rate of 96％ for the acoustic emission signals of PMIA ( poly-m-phenylene isophthalamide) and high performance vinylon fiber.%针对纤维拉伸断裂声发射信号的非平稳性、信号特征间高度重叠等问题,提出一种声发射信号特征提取及纤维种类诊断的模型,可用于识别拉伸断裂的纤维种类.首先,通过小波变换、增强经验模态分解方法(EEMD)对不同种纤维的拉伸断裂声发射信号进行预处理、分解;然后,结合主成分分析法(PCA)提取频率特征;最后,运用最小二乘支持向量机(LSSVM)对纤维拉伸断裂的特征频率进行分类识别.结果表明,主成分分析法可将信号特征降维,并降低不同纤维频率特征之间的相关性,提高了对纤维拉伸断裂声发射信号的准确识别.针对芳纶1313、高性能维纶纤维拉伸断裂的声发射信号,EEMD-PCA-LSSVM模型的总识别率达96％.