摘 要：针对刀具磨损监测中信号的非平稳特性和小样本建模中神经网络容易陷入局部值的问题，提出基于多传感器信号，运用集合经验模态分解（ensemble empirical mode decomposition，EEMD）和支持向量机（support vector machine，SVM）相结合的算法，实现对刀具磨损多状态的识别。首先对振动信号进行集合经验模态分解，将其分解为若干个本征模态函数（intrinsic mode function，IMF）之和，然后计算得到三向切削力信号的均值和各本征模态函数分量的能量百分比值作为磨损状态分类特征，最后运用支持向量机和Elman神经网络对刀具在不同磨损状态下的特征数据样本进行训练和识别。实验结果证明该方法能很好地实现对刀具磨损状态的识别，与Elman神经网络相比，支持向量机具有更高的识别率，更适合小样本情况下刀具磨损状态的分类识别。
Study of tool wear based on EEMD-SVM
JIANG Yan， FU Pan， LI Xiaohui
（School of Mechanical Engineering，Southwest Jiaotong University，Chengdu 610031，China）
Abstract: To make the signals steady in cutting-tool wear monitoring and prevent neural networks from easily falling into local minimum values during small sample modeling， we have proposed a new method to identify cutting-tool wear conditions based on multi-sensor signals， ensemble empirical mode decomposition（EEMD） and support vector machine（SVM）. First， collected vibration signals are decomposed into a number of stationary intrinsic mode functions and further into the sum of multiple intrinsic mode functions. Second， these functions are used to calculate the mean value of three-direction cutting force signals and the energy percentage of each intrinsic mode function component and the calculation results were taken as the classification features of wear conditions. Next， the characteristic samples under different wear extents were trained and identified by SVM and Elman Neural Network. The experiment shows that this method can be used to determine the wear conditions of cutting tools and the SVM has a higher identification rate and more suitable for classified identification of cutting-tool wear conditions for small samples.
Keywords: tool wear condition identification； ensemble empirical mode decomposition； support vector machine； multi-sensor