【作者】 唐亮；

【导师】 张洁；

【作者基本信息】 西南交通大学 ， 测试计量技术及仪器， 2014， 硕士

【摘要】 机械制造业直接体现一个国家的科技水平和经济实力,同时也是其他众多行业新技术、新产品的研发和生产平台,在国民经济发展中有着重要的地位。数控刀具直接参与机械的制造,且其运行状态对加工精度以及加工连续性至关重要,使刀具运行状态监测技术成为发展现代制造技术的关键。本论文以先进的多传感器信息融合技术为基础,搭建了以刀具振动信号以及刀具切削力信号为监测信号的刀具磨损状态监测平台。通过连续切削以及重复试验,采集刀具从新刀到严重磨损整个过程的振动以及切屑力信号参数,并以此为基础展开刀具状态监测技术的研究。针对现场采集的数据因为干扰会包含噪声的问题,本文引入匹配追踪算法,通过该方法对原始信号进行自适应的分解重构,达到了滤波的目的,提高了信号的信噪比；结合时域、频域以及基于小波包时频域分析方法,提取了监测信号的特征参数；针对输入数据样本量大以及样本维数较高的问题,通过核主成分分析(KPCA)方法,在损失状态信息较少的情况下,将多个特征指标转化为几个综合的特征指标,用较少的特征参数来代表轴承状态的绝大部分信息,实现了特征参数的降维。针对BP神经网络在训练时,网络收敛速度慢,易陷入局部最小值,甚至导致网络无法完成训练等问题,引入人工蜂群算法对BP神经网络的性能进行优化。该算法中引入的种群概念,以及算法中种群的进化机制,大大提升了BP神经网络的性能。实验数据表明,优化后的网络克服了原网络易陷入局部最小值的缺陷,同时提高了网络对刀具状态监测的精度,实现了刀具磨损状态的分类识别。这些研究对不断完善刀具状态监测技术具有一定的实际意义。更多还原

【Abstract】 Machinery manufacturing industry directly reflects the level of technology and economic of a country. It’s also the platform for new technology development and new product production of many other industries. And it plays an important role in national economic development. CNC tools are involved in manufacturing of machinery directly, and its operational status is essential for precision and continuity of machining. So the tool condition monitoring technology is the key to develop modern manufacturing technology.In this paper, tool wear condition monitoring use tool vibration and cutting force signals as the monitoring signal which based on advanced multi-sensor fusion technology. Collecting vibration and cutting force signals of cut tools through new knife to severe wear by continuous cutting and repeated tests, which is used for studying tool condition monitoring techniques.For the signals collected in the field will contain noise signal which come from interference. This paper introduces matching pursuit algorithm, we can use it to make the adaptive decomposition and reconstruction of original signals, which can realize the purpose of filtering and improve the signal to noise ratio; Using analysis methods including time domain, frequency domain and time-frequency domain which based on wavelet packet can extract the parameters of the monitoring signals. For the problem of input data sample which has big volume and high dimensionality, Kernel Principal Component Analysis (KPCA) method can transform multiple features into several intergrated characteristics with less information lost and uses few characteristic parameters to represent the major information and achieve dimensionality reduction of characteristic parameters.Artificial bee colony algorithm is introduced to optimize BP neural network performance for the BP neural network during training having problems like slow convergence speed, falling into local minimum easily, and even leading to the network can not complete the training and other issues. The algorithm introduced the concept of population, as well as the mechanism of evolution algorithm population, which greatly enhance the performance of BP neural network.The data of the research shows that optimized BP neural network has overcome the defect of original network for always easily falling into local minimum and improves the accuracy of tool condition monitoring. All of this is practical and significant to develop cutting tools condition monitoring techniques.更多还原