【作者】 刘芽；

【导师】 傅攀；

【作者基本信息】 西南交通大学 ， 仪器仪表工程， 2012， 硕士

【摘要】 本文针对机械加工中大量用到的切削刀具的磨损状态进行了监测方法的研究,主要做了以下一系列工作：首先,在分析刀具损坏机理和刀具磨损过程以及刀具磨钝标准的基础上,搭建了融合多传感器的刀具磨损状态监测实验平台,并且采集车削时不同切削条件下的振动和切削力数据。其次,详细分析经验模态分解方法的原理和分解过程,针对其存在的模态混叠不足采用改进的总体经验模态分解方法。尝试性的将该方法运用于刀具磨损振动数据的分析处理上,抽取经过分解后的各IMF分量的能量百分比值作为表征刀具磨损量的特征值,并且详细分析证明了所提取特征值具有很好的重复性和差异性。此外,介绍了统计知识基础和基于结构风险最小化原理的支持向量机的基本思想、理论特点。将运用EEMD分解法提取的振动数据能量百分比值和切削力均值作为输入支持向量机的训练样本和测试样本,建立支持向量机分类器模型,经过测试样本的验证发现支持向量机能够对EEMD能量百分比特征值进行正确的分类识别。最后,经过对比BP和RBF神经网络与支持向量机分类器对刀具磨损状态的识别,发现支持向量机在识别精度、训练时间和对模型结构的依赖程度等方面表现出很好的优越性。本文将EEMD分解方法与支持向量机结合起来运用到刀具磨损状态监测中,得到预想的结果,丰富了刀具磨损状态监测的研究方法,也为进一步实现在线监测提供一种理论依据。此外还对比分析了神经网络和支持向量机在分类识别应用中的特点,为后续的识别模型选取提供一定的理论参考。更多还原

【Abstract】 In this paper, we have a series of work for the method research for the wear state of cutting tool, which is been used in machining largely.Firstly, the mechanism of tool damage、the process of tool wearing and the tool blunt standard are been analyzed. Then setting up the experimental platform of multi-sensor for the monitoring of tool wear condition, and collecting turning vibration and cutting force data under different conditions.Secondly, author analysis the principle of the method of empirical mode decomposition and the decomposition process. Towards its shortage of modal aliasing, it used the improved overall empirical mode decomposition method. Then trying to use the method for the analysis of the vibration data of tool wear. It extracted the percentage value of the energy of each IMF component after decomposition as the feature, and discussed the characteristic’s repeatability and differences in detail.In addition, it introduced the basic idea of the statistical knowledge and the theoretical characteristics of support vector machine. Support vector machine is based on structural risk minimization principle. Then, it makes the energy percentage values of vibration data extracted from ensemble empirical mode decomposition and the mean of cutting force as training samples and testing samples of support vector machine model, In this way, it established a support vector machine classifier model. The test samples tell the correct of this model for classification and identification of feature value that the energy percentage values of vibration data extracted from ensemble empirical mode decomposition and the mean of cutting force.Finally, though comparing of BP and RBF neural network with support vector machine classifier in tool wear identification found that support vector machine showed good superiority in recognition accuracy, training time and reliance on the model structure. This paper combined EEMD decomposition method and support vector machine for tool wear monitoring, and achieved the anticipated results. Enriched the research methods of the tool wear condition monitoring, and also provided a theoretical basis for the further realization of online monitoring, In addition, it compared the characteristics of neural network and support vector machine in classification and recognition applications for providing a theoretical reference for the subsequent selecting of recognition model.更多还原