多源声发射信号混合重叠组稀疏分类研究

Research on Mixed Overlapping Group Sparse for Multi-source Acoustic Emission Signal

针对高速列车车体裂纹声发射检测的多源、波模式重叠及噪声干扰问题,提出一种基于本征模态的混合重叠组稀疏(MOGS)分类方法用于声发射源识别。MOGS是一种兼顾组间和组内稀疏,同时允许类间特征重叠的结构稀疏模型。设计了一种新的噪声预分解矩阵以降低本征模态分解计算量,选取目标特征频带模态为分类样本来提高类间差异。通过K-SVD层次稀疏组套索罚训练MOGS类别字典,并给出一种罚函数块坐标可分离的近似光滑处理过程以实现MOGS套索求解。实验表明,该方法对几类多源含噪信号分类准确率均高于80%,在识别率和波形重构效果上优于对比方法。

Acoustic emission detection of high-speed train body cracks involving multiple sources,overlapping wave modes,and noise interference,a mixed overlapping group sparse(MOGS)classification method with intrinsic mode function(IMF)is proposed for the identification of acoustic emission sources.MOGS is a structured sparse model that involve inter-and intra-group sparsity while allowing feature overlap between classes.A new noise pre-decomposition matrix is designed to reduce the computational complexity of IMFs.The IMF that include eigenfrequencies was selected as the sample to improve the difference between classes.MOGS dictionary was trained by the K-SVD with hierarchical group sparse lasso penalty function,and a separable block coordinates with approximate smoothing process method was proposed to solve MOGS lasso penalty function.Experiments show that the classification accuracy of this method is higher than 80%,the identification rate and waveform reconstruction effect are better than other algorithms.