基于多类超球支持向量机的铣削颤振预测方法

为了预测薄壁件铣削过程颤振的发生,提出了一种应用小波系数特征和多类超球支持向量机进行铣削颤振预报的方法.首先基于连续小波变换分别提取高、低频段铣削振动信号的特征,然后利用多类超球支持向量机进行稳定铣削状态、铣削颤振孕育状态、铣削颤振状态识别.为了简化支持向量机进行多类分类时所带来的计算复杂性,该算法使每一类样本都获得一个超球支持向量机,在特征空间中以测试样本与超球中心距离、超球半径作为决策函数来进行识别.实验表明,在铣削颤振识别系统中多类双核超球支持向量机与连续小波系数特征向量相结合具有良好的识别效果,颤振孕育预报正确率达98.0％.     

基于粒子群优化的VB-LSSVM算法研究辛烷值预测建模

针对现有红外线分析仪表无法实现阶段在线检测车用汽油调合中,MMT抗爆剂对辛烷值的影响问题,考虑到样本数据较少的因素,提出一种基于粒子群优化算法的矢量基最小二乘支持向量机方法,首先以粒子群优化的方法来选取最小二乘支持向量机的模型参数,然后用矢量基判据选择支持向量,使最小二乘支持向量机的解具有稀疏性.该方法不但克服了常用的交叉验证法的耗时与盲目性问题,发挥了最小二乘支持向量机的小样本学习和计算简单的特点,而且提高了最小二乘支持向量机模型的泛化能力,将其应用于汽油调合系统中研究法辛烷值的预测,仿真结果表明,该方法是可行且有效的.     

基于超球面支持向量机的丝杠故障诊断技术

针对高档数控机床丝杠故障样本不易获取以及样本分布不均的问题,提出了一种用小波包分解和超球面支持向量机进行分类的丝杠故障智能诊断技术.该方法将振动信号小波包分解后的频带能量作为特征向量,输入到超球面支持向量机分类器进行故障识别.通过改变相关参数,研究了模型参数选择在构造超球面支持向量机中的重要作用.试验结果表明,建立的超球面支持向量机模型能够有效地对机床丝杠故障进行诊断.     

异物在线识别中一类支持向量机机理及实现

针对高速异物在线识别中正常物料与异物颜色差异及异物颜色随机性的特点，研究了一类支持向量机在异物识别中的特性，提出了一种基于一类超球面支持向量机的在线识别算法，通过OC-SVM确定正常物料的颜色分布，从而对异物进行识别。在求解OCSVM过程中，提出了超球面离心系数ω的概念，并采用Zoutendijk可行方向机制确定工作集，简化了序列最小优化算法。研究工作表明，该算法速度快，整体运算时间比LibSVM减少20%；识别率高，尤其对与正常物料颜色接近的异物有明显效果，与一维和三维正态拟合算法相比，整体识别率提高约8～10%。     

核极化的多核LSSVM及其在分类中的应用

为了解决最小二乘支持向量机对于选择核函数盲目性的问题,将核度量标准核极化和多核学习引入最小二乘支持向量机中,提出了基于核极化的多核最小二乘支持向量机算法。算法首先利用核极化确定每个基本核函数的权系数,再根据多核学习原理组合多核函数,然后,建立多核最小二乘支持向量机模型,并进行模型的学习训练和预测。UCI数据上的试验结果表明,所提出的算法比SVM、最小二乘支持向量机和其他的多核学习方法具有更高的分类准确率。     

基于连续小波和多类球支持向量机的颤振预报

研究了一种应用连续小波特征和多类球支持向量机进行铣削系统颤振预报的方法,该方法基于连续小波变换提取铣削振动信号的特征,利用多类球支持向量机对正常铣削状态、颤振孕育状态和颤振爆发状态的振动信号进行三分类识别,通过识别颤振孕育状态预测颤振爆发。试验结果表明,在铣削颤振识别与预测中,铣削振动信号的连续小波特征与多类球支持向量机相结合具有良好的识别颤振孕育状态和颤振爆发状态的能力,颤振孕育状态的识别正确率达95.0%,颤振爆发状态的识别正确率达97.5%。     

油液在线监测系统中磨粒识别技术研究

针对磨损状态监测要求,构建了基于显微图像分析的油液在线监测系统。根据系统光路特点,对磨粒图像进行了基于彩色特征的转换,并通过与背景图像的差值处理来快速提取磨粒目标。基于最小二乘支持向量机设计了磨粒两类分类器,并利用粒子群优化算法对最小二乘支持向量机模型中的参数进行了优化选取;根据磨粒识别体系,设计了基于最小二乘支持向量机的磨粒综合分类器。最后,利用铁谱分析技术对系统性能和识别效果进行了检验,结果表明本系统具有较高的检测精度和识别效果。     

基于多最小二乘支持向量机的草酸钴粒度软测量

提出了一种基于改进的鲁棒学习方法(improved robust learning algorithm,IRLA)的多最小二乘支持向量机(multipleleast squares support vector machine,Multi-LSSVM)建模方法,用以解决非线性系统建模问题。该方法通过Bootstrap算法复制出训练集样本空间上的多个样本子空间,训练出多个成员最小二乘支持向量机模型,然后应用改进的鲁棒学习方法对成员最小二乘支持向量机模型的权重进行优化融合,从而使多最小二乘支持向量机模型具有较高的准确率和泛化能力。通过仿真实验,验证了方法的有效性;并将其应用于湿法冶金合成过程草酸钴粒度软测量建模问题,获得了比单个最小二乘支持向量机模型方法更高的预测精度。     

基于MSVM的多品种小批量动态过程在线质量智能诊断

提出了基于多分类支持向量机(MSVM)的多品种、小批量动态过程在线质量智能诊断方法。离线训练时,提取异常模式仿真数据的小波重构特征,对 MSVM识别和估计模型进行训练和测试,同时建立异常因素诊断库;在线诊断时,对“监控窗口”数据特征的过程模式及参数进行识别与估计,而后利用异常因素诊断库实现对多品种、小批量动态过程实时在线智能诊断。某精密轴加工过程实例验证了该智能诊断方法的有效性。     

基于最小二乘支持向量机的电力变压器故障诊断

文章提出使用最小二乘支持向量机(LS-SVM)作为分层决策电力变压器故障诊断模型.首先根据DGA技术以及相关统计分析,选择典型油中故障气体的相对含量作为特征量,然后利用数值预处理后得到的数据样本分别对四级最小支持向量机分类器进行训练和识别,并最后判断输出变压器所处的状态,且针对最小二乘支持向量机存在的参数选择问题,使用了多层动态自适应优化算法来优化最小二乘支持向量机参数.仿真结果表明LS-SVM是一种较为有效的非线性建模方法,具有较快的收敛速度和较高的计算精度,满足电力变压器故障诊断的要求.     

Hybrid model- and signal-based chatter detection in the milling process

Chatter causes machining instability and reduces productivity in the metal cutting process. It has negative effects on the surface finish, dimensional accuracy, tool life and machine life. Chatter identification is therefore necessary to control, prevent, or eliminate chatter and to determine the stable machining condition. Previous studies of chatter detection used either model-based or signal-based methods, and each of them has its drawback. Model-based methods use cutting dynamics to develop stability lobe diagram to predict the occurrence of chatter, but the off-line stability estimation couldn’t detect chatter in real time. Signal-based methods apply mostly Fourier analysis to the cutting or vibration signals to identify chatter, but they are heuristic methods and do not consider the cutting dynamics. In this study, the model-based and signal-based chatter detection methods were thoroughly investigated. As a result, a hybrid model- and signal-based chatter detection method was proposed. By analyzing the residual between the force measurement and the output of the cutting force model, milling chatter could be detected and identified efficiently during the milling process.

     

切削加工颤振智能监控技术

切削加工颤振智能监控技术是智能机床中不可或缺的一部分,是智能加工的一个重要发展方向。它对于提高零件的加工精度与效率,增加企业的运营绩效具有重要的意义。以传感器的选择、特征提取、颤振识别和颤振抑制为主线,系统的综述了切削加工过程中颤振智能监控的研究进展。分析颤振信号的选择和时域、频域、时频域以及特征自适应智能提取的特征提取方法;分析神经网络、支持向量机、隐马尔科夫模型、混合模型和在线智能进化模型在颤振识别中的应用;着重分析基于主轴转速调整的颤振智能控制方法。在此基础上,对切削加工颤振智能监控的研究难点进行了分析,并总结了目前存在的问题。最后,对切削加工颤振智能监控技术今后的发展趋势进行了展望。     

镗削颤振快速预报技术研究

切削颤振切导致产品质量、生产效率、刀具和机床使用寿命的降低 ,同时造成噪声污染 ,影响操作者身心健康。随着工厂自动化发展 ,操作者日益远离加工现场 ,对切削颤振进行在线监视预报和控制变得越来越重要。但由于其发生发展的过程极其短暂 ,使颤振的在线预报十分困难。本文在讨论了镗削颤振发展机理的基础上 ,提出了一种基于神经网络进行颤振预报的新方法。用 L O- RBF模型进行传感信号预处理 ,结果输入 Fuzzy ARTMap模型进行颤振识别 ,大大缩短了信号处理时间 ,提高了识别的准确性 ,取得了满意的结果。     

Chatter reliability prediction of side milling aero-engine blisk

Reliability analysis of a dynamic structural system is applied to predict chatter of side milling system for machining blisk. Chatter reliability is defined as the probability of stability for processing. A reliability model of chatter was developed to forecast chatter vibration of side milling, where structure parameters and spindle speed are regarded as random variables and chatter frequency is considered as intermediate variable. The first-order second-moment method was used to work out the side milling system reliability model. Reliability lobe diagram (RLD) was applied to distinguish reliable regions of chatter instead of stability lobe diagram (SLD). One example is used to validate the effectiveness of the proposed method and compare with the Monte Carlo method. The results of the two approaches were consistent. Chatter reliability and RLD could be used to determine the probability of stability of side milling.

     

Chatter stability prediction for high-speed milling through a novel experimental-analytical approach

Chatter prediction is crucial in high-speed milling, since at high speed, a significant increase of productivity can be achieved by selecting optimal set of chatter-free cutting parameters. However, chatter predictive models show reduced accuracy at high speed due to machine dynamics, acquired in stationary condition (i.e., without spindle rotating), but changing with spindle speed. This paper proposes a hybrid experimental-analytical approach to identify tool-tip frequency response functions during cutting operations, with the aim of improving chatter prediction at high speed. The method is composed of an efficient test and an analytical identification technique based on the inversion of chatter predictive model. The proposed technique requires few cutting tests and a microphone to calculate speed-dependent chatter stability in a wide range of spindle speed, without the need of stationary frequency response function (FRF) identification. Numerical and experimental validations are presented to show the method implementation and assess its accuracy. As proven in the paper, computed speed-dependent tool-tip FRF in a specific configuration (i.e., slotting) can be used to predict chatter occurrence in any other conditions with the same tool.     

基于内置力执行器的铣削颤振的主动控制

高速加工中铣削颤振不仅降低工件的表面加工质量,严重时还会造成刀具或者其他加工部件的损坏,因此对电主轴铣削颤振进行控制具有重要的意义。为对电主轴铣削过程中的颤振进行有效控制,在双绕组无轴承感应电动机的基础上,提出一种具有内置力执行器的感应型高速电主轴结构,建立电主轴—刀具系统的有限元模型、动态铣削模型、双绕组感应型电主轴电磁力模型,在对具有内置力执行器的感应型高速电主轴电磁力进行解耦后,提出基于内置力执行器的电主轴铣削颤振的主动控制方案,通过仿真分析控制器的主要参数对电主轴铣削稳定性的影响。结果表明采用具有内置力执行器的感应型高速电主轴能够有效地提高电主轴铣削的稳定区域以及在抑制铣削颤振方面具有明显效果。     

基于HE-SVDD的航空发动机工作状态识别*

针对人工识别航空发动机工作状态的复杂性和耗时性,提出一种基于超椭球分类面支持向量数据描述(HESVDD)的快速识别方法。首先构建了一个根据训练样本分布特征可调的HE-SVDD分类器,使之具有从大规模飞行数据中快速识别发动机工作状态的能力;然后研究了航空发动机状态识别的参数选取和样本生成问题;最后采用HE-SVDD对两个飞行架次的发动机工作状态进行了识别。结果表明,该方法能快速准确地识别出发动机的工作状态,可应用于发动机状态的在线或离线监控。     

Image processing for chatter identification in machining processes

Identifying chatter or intensive self-excited relative tool–workpiece vibration is one of the main challenges in the realization of automatic machining processes. Chatter is undesirable because it causes poor surface finish and machining accuracy, as well as reducing tool life. The identification of chatter is performed by evaluating the surface roughness of a turned workpiece undergoing chatter and chatter-free processes. In this paper, an image-processing approach for the identification of chatter vibration in a turning process was investigated. Chatter is identified by first establishing the correlation between the surface roughness and the level of vibration or chatter in the turning process. Images from chatter-free and chatter-rich turning processes are analyzed. Several quantification parameters are utilized to differentiate between chatter and chatter-free processes. The arithmetic average of gray level G _a is computed. Intensity histograms are constructed and then the variance, mean, and optical roughness parameter of the intensity distributions are calculated. The surface texture analysis is carried out on the images using a second-order histogram or co-occurrence matrix of the images. Analysis is performed to investigate the ability of each technique to differentiate between a chatter-rich and a chatter-free process. Finally, a machine vision system is proposed to identify the presence of chatter vibration in a turning process.     

Improved experimental-analytical approach to compute speed-varying tool-tip FRF

Chatter stability prediction is crucial to improve the performances of modern milling process, and it gets even more important at high speeds, for which very productive cutting parameters can be achieved if the suitable spindle speed is selected. Unfortunately, the available chatter predictive models suffer from reduced accuracy at high speed due to inaccuracies in the input data, especially the machine tool dynamics that is acquired in stationary configurations but could sensibly change with spindle speed. In this paper, an efficient method to identify the speed-varying Frequency Response Functions (FRFs) under operational conditions is presented. The proposed approach is based on the definition of some experimental chatter limits (i.e., chatter frequency and related depth of cut), obtained by a dedicated test, called Spindle Speed Ramp-up. The experimental results are then combined with the analytical stability solution. By minimizing the differences between the experimental and predicted chatter conditions, a dedicated algorithm computes the speed-varying FRFs. Few tests and simple equipment (i.e., microphone) are enough to calculate the FRFs in a wide range of spindle speeds. The proposed technique was validated in real machining applications, the identified tool-tip FRFs are in accordance with expected trend reported in scientific literature. Speed-varying stability lobe diagram reconstructed with the computed FRFs is proven to be accurate in predicting stable cutting parameters.