基于局部线性嵌入的特征融合方法在岩石破裂 状态分类的应用*

为了获取岩石破裂过程有效的声发射信号特征,更好的对岩石破裂状态进行分类,提出一种基于流形学习算法的LLE特征融合方法进行数据降维。以红砂岩为研究对象设计室内单轴压缩实验采集信号,然后对原始声发射信号预处理并对信号进行特征提取,以时域、频域下的特征向量重新组合成一组新的多维特征向量,采用线性主元(PCA)和流形学习LLE算法分别进行降维。比较两种算法降维后融合特征的聚类效果二维和三维分布图,使用LLE算法降维后,四种状态分布相对更近,呈一条水平线趋势,且各状态交叉混叠数目较少,第一状态没有一个样本错判,且四个状态相比于PCA降维后的聚类效果更集中。再比较两种算法降维后融合特征的敏感度之和,LLE算法融合特征敏感度之和远大于PCA算法,说明经过LLE算法降维后得到的融合特征更多地表征了原始信号包含的局部信息同时证明了LLE算法相比PCA算法具有更好的聚类效果。最后经LLE特征融合下的砂岩破裂状态分类实验验证,融合特征后的识别率相对单一的时域特征识别提高了6%。表明该方法能显著提高岩石破裂状态分类的识别率,降维性能相对突出。

Application of locally linear embedding-based feature fusion method in rock fracture state classification

In order to obtain the effective AE signal characteristics of the rock fracture process and better classify the rock fracture state, A LLE feature fusion method is presented for data dimension reduction. With red sandstone as the research object design indoor uniaxial compression experiment to collect the signal, and then the original acoustic emission signal pretreatment and signal features extracted, the feature vectors in the time and frequency domains are recombined into a new set of multi-dimensional feature vectors, using linear principal (PCA) and manifold learning LLE algorithm respectively.Comparing the two algorithms after the fusion effect of two-dimensional and three-dimensional distribution, using LLE algorithm, four state distribution is relatively closer, showed a horizontal line trend, and each state cross aliasing number is less, the first state without a sample error, and four states compared with the PCA after the clustering effect is more concentrated.By comparing the sum of the sensitivity of fusion features of the two algorithms after dimension reduction, the sum of fusion features of LLE algorithm is much larger than that of PCA algorithm, which shows that the fusion features obtained by LLE algorithm after dimension reduction represent more of the local information contained in the original signal and proves that LLE algorithm has better clustering effect compared with PCA algorithm.Finally, according to the classification of sandstone fracture states under LLE feature fusion, the recognition rate of the fusion features increased by 6% compared with a single time-domain feature identification.It shows that this method can significantly improve the identification rate of rock rupture state classification, and the dimension reduction performance is relatively outstanding.