基于最小二乘支持向量机对刀具切削状态的识别

基于小波包优良的时频特性和最小二乘支持向量机(least squares support vector machine,LS-SVM)对于小样本出色的学习泛化能力,提出了一种研究刀具切削状态的方法.采用最小熵准则对声发射信号进行最佳小波包分解,以各频段的信号能量占总能量的百分比来构造特征向量,输入LS-SVM多类分类器,实现对刀具切削状态的分类识别.实验结果表明,在采用高斯核函数的LS-SVM多分类算法中,选取惩罚因子γ=10,径向基核参数σ2=1时,该分类器能对测试样本进行准确的刀具切削状态识别.

Recognition of Tool Cutting State Under Least Squares Support Vector Machine

A method of cutting tool condition based on wavelet packet excellent time-frequency characteristics and least squares support vector machine(LS-SVM) high-quality learning and generalization ability with small samples is presented for a cutting tool state recognition system.Minimum entropy criterion was adopted to decompose best wavelet packet for extracting feature of acoustic emission signals,the feature vectors were constructed by the AE signals energy relative percentage of each band accounted for the total energy,which were brought in multi-class LS-SVM classifier,and the classification recognition of different cutting tool states was achieved.Results show that the multi-class LS-SVM classifier is an efficient method for accurately recognizing the cutting tool states of the test samples that contain feature vectors,when γ=10(penalty factor) and σ2=1(RBF kernel parameter) in the LS-SVM multi-classification algorithm with Gaussian kernel function.