【作者】 朱丹丹；

【导师】 陈永洁；

【作者基本信息】 华中科技大学 ， 机械电子工程， 2005， 硕士

【摘要】 鉴于前人的研究成果,本文较为系统、全面的介绍了切屑控制技术的概念和发展状况,阐述了对可转位三维槽型刀片断屑范围进行预测的目的和意义,在分析现代断屑范围预测技术发展的基础上,探讨了这种技术在实现过程中存在的难题和不足,提出了利用RBF 神经网络对可转位三维槽型刀片的断屑范围进行预测。RBF 网络是前馈型神经元网络,它通过改变神经元非线性变换函数的参数来实现非线性映射,具有学习速度快等优点,近年来在自动控制等众多领域受到重视。本文在Matlab 环境下实现了最近邻聚类学习算法,为建立了预测模型奠定了基础。设计网络学习样本的过程中,我们进行了充分的切削试验,深入研究了加工条件和刀片几何特征对切屑控制的影响,在此基础上提出了将三维断屑槽按截面分类,然后从典型截面中提取最有代表性的特征并将其数值化的方法,并将其应用于实践之中,此方法为如何全面而简洁的提取断屑槽几何特征提供了一种有效的途径。同时也为如何确定神经网络的输入参数提供了一种简易可行的方法。在上述基础上,本文利用RBF 神经网络对可转位三维槽型刀片断屑范围进行预测,并在试验中取得良好的效果。通过分析各网络参数在RBF 神经网络实现预测功能的过程中的影响,提出了完善学习样本和改进神经网络结构的设计来提高所设计的神经网络的学习和泛化能力,从而实现较高的识别率,进一步拓宽了切屑控制预测的研究方法。更多还原

【Abstract】 Based on the achievement that previous scholars make, the conception and development of chip control technology are systematically introduced in the dissertation. The significance of the prediction of the breaking area of 3-D chip former is clarified. The weakness of the prediction of the breaking area of 3-D chip former is also analyzed. Finally the new method of the prediction of the breaking area using neural networks for 3-D chip former is presented. The Radius Basis Function (RBF) networks, featuring a quick learning speed, is one of the feedforward neural networks. It has attracted much attention from automatic control recently for it implementing the nonlinearity mapping through adjusting the parameters of the nonlinearity transfer function. One of the on-line adaptive learning algorithm—the nearest neighbor-clustering algorithm has been implemented in the MATLAB programming language in this dissertation, laying the foundation for the prediction model. During the course of designing the learning samples, we make a lot of cutting experiments, after penetratingly studying the effect of the machining condition and the geometry feature of the chip former on chip control, the dissertation presents an improving method. First, classifying the chip-breaking grooves according to their sectional feature; then selecting the typical features from several important sections; lastly, expressing these features in the form of numberical value. This method is useful for effectively selecting the geometry feature of the chip former. Based on the above, this dissertation come true the prediction of the breaking area of 3-D chip former based on RBF neural networks and the result is good. After studying the effect of the network parameters, we get the conclusion that optimizing the learning samples and improving the network structure are benefit to training and advancing generalization capability of neural network, this way farther develops the research method of predicting the chip control.更多还原