基于双谱奇异值分解的刀具磨损识别方法

Identification Method for Tool Wear Based on Bispectrum Singular Value Decomposition

针对刀具磨损声发射信号的非线性、非平稳特性,提出一种基于双谱奇异值分解的刀具磨损特征提取方法。对刀具不同磨损阶段的声发射信号进行双谱分析,构造初始特征向量矩阵,然后对初始特征向量矩阵进行奇异值分解,计算奇异谱,将奇异谱作为刀具磨损特征向量,利用最小二乘支持向量机对刀具磨损状态进行识别。实验结果表明:所提取的特征可以很好地反映刀具的磨损状态,最小二乘支持向量机更适于在小样本下实现刀具磨损状态的识别,与神经网络识别方法相比具有更高的识别率。

In view of the nonlinear and non-stationary characteristics of acoustic emission signal of tool wear, a feature extraction method based on bispectrum singular value decomposition was proposed. The bispectrum analysis method was used to decompose the collected acoustic emission signals that reflecting the different tool wear stage, the initial feature vector matrix was constructed. Then u- sing initial feature vector matrix, the singular spectrum was calculated by the method of singular value decomposition. Least squares support vector machine （ LS-SVM ） was selected to identify the tools wear state. The identification result shows that the bispectrum fea- ture can better reflect the tool wear state. LS-SVM is efficient, feasible and superior to neural network, and it has a higher identifica- tion rate.