摘要
针对刀具磨损过程中产生的非平稳性信号,提出了基于变分模态分解的关联维数及相关向量机的刀具磨损状态监测方法。首先,利用变分模态分解对采集的声发射信号进行分解,获得一系列分量;其中部分分量跟磨损状态相关,部分分量是干扰噪声。为此根据分解后分量与原信号的互信息值提取出敏感分量;利用刀具信号特点确定关联维数的时延参数和嵌入维数,计算敏感分量的关联维数并组成特征向量;最后,将刀具不同状态的特征向量输入相关向量机进行训练与测试,从而实现对刀具磨损状态的监测。实验结果表明,该方法能够有效地识别出刀具磨损过程中不同的工作状态,且分类准确率较经验模态分解好。
According to unstable-state and non-linear characteristics of tool wear signal, a recognition method based on variational mode decomposition correlation dimension and relevance vector machine is proposed. First, the acoustic emission signal was decomposed by variational mode decomposition, then a series of components were obtained. The components generated by variational mode decomposition have different sensitivity to condition of tool wear. According to the mutual information of components, sensitive components were selected, which were used to calculate the correlation dimension of sensitive components and combined into a feature vector. Finally, the feature vector were input relevance vector machine, which will be classified, trained and tested, in order to identify the state of tool wear. By comparing classification accurate rate of variational mode decomposition and applied empirical model decomposition methods, the superiority of the proposed method based on variational mode decomposition is demonstrated in state recognition of tool wear.
出处
《计量学报》
CSCD
北大核心
2018年第2期182-186,共5页
Acta Metrologica Sinica
基金
国家自然科学基金(51475158)
国家科技重大专项(2012ZX04003041)
关键词
计量学
刀具磨损
声发射法
状态识别
变分模态分解
关联维数
相关向量机
metrology
tool wear
acoustic emission method
state recognition
VMD
correlation dimension
relevance vector machine
