针对铣削刀具磨损状态识别问题,提出谐波小波包和最小二乘支持向量机(LS-SVM)的状态识别方法.为克服传统小波包分解的频带交叠问题,采用谐波小波包提取不同磨损状态下铣削力信号的各频段信号能量,归一化处理后,输入LS-SVM多类分类器,实现铣削刀具磨损状态的识别.针对LS-SVM的惩罚因子和核参数对模型识别精度影响较大的问题,提出回溯搜索算法(BSA)进行自动参数寻优.实验结果表明,谐波小波包比小波包在刀具磨损状态特征提取时具有更好的识别效果.与粒子群算法进行比较,证明BSA优化LS-SVM具有更高的识别精度.%Aiming at the problems of milling tool wear state recognitions,a state recognition method was proposed based on harmonic wavelet packet and LS-SVM.To overcome the band overlap-ping problems in traditional wavelet packet decompositions,the milling force signal energies of each bands were extracted in different wear states by harmonic wavelet packet,which were brought in multi-class LS-SVM classifier after normalizing,then the classification recognition of different cutting tool states was achieved.BSA was proposed to search the optimal values of the kernel functional pa-rameters and error penalty factors which affected the precision of the LS-SVM significantly.Experi-mental results show that harmonic wavelet packet is more effective and feasible than wavelet packet, and the proposed milling tool wear recognition method has higher accuracy.
展开▼