基于多元相空间重构的数控机床运动精度预测

发布时间：2018-04-21 06:28

本文选题：数控机床 + 运动精度　；参考：《重庆理工大学》2017年硕士论文

【摘要】：精度是数控机床的一个重要性能指标。数控机床精度直接影响加工零件的精度和质量,体现着整个制造业的技术水平和竞争力。我国是数控机床消费大国,但不是数控机床制造强国,需求和制造力处于不平衡状态,尤其是高档数控机床。与进口数控机床相比,国产数控机床的精度保持性不高,提高精度保持性是国内数控机床亟待解决的问题。本文针对数控机床运动精度进行了分析和预测,根据分析和预测结果,可在数控机床运动精度失效之前采取措施减小或消除误差来源,从而提高产品加工精度,并延长数控机床的服役时间,提高精度保持性。另外还能指导机床维修工作,防止失修带来的经济损失。本文针对数控机床的多个运动精度特征量时间序列,基于多元相空间技术建立运动精度相空间,将低维时间序列映射到高维相空间中,恢复混沌吸引子,揭露复杂表象后的有序状态。引入主成分分析法对重构后的相空间进行去冗降维,简化模型结构。然后构建小波神经网络预测模型,以运动精度的相点坐标为输入,圆度误差变化量为输出,对预测模型训练学习,从而实现数控机床运动精度的预测。本文主要的研究工作如下:(1)搭建实验平台测试数控机床运动精度,获取多个运动精度特征量时间序列数据。并通过算数平均法对数据进行降噪处理,采用最大李雅普诺夫指数分析数控机床运动精度序列的混沌特性。(2)重构数控机床运动精度相空间。采用C-C算法分别计算出各运动精度特征量时间序列的相空间重构参数。然后通过主成分分析法滤除冗余信息,简化相空间结构。以运动精度相空间维数作为小波神经网络输入层神经元数目,避免了试凑法,保证了输入信息的完备性。(3)构建数控机床运动精度小波神经网络预测模型。选用Morlet小波作为隐含层神经元激励函数,分析并确定小波神经网络结构,给出了预测模型参数修正算法,然后通过重构后的运动精度相空间数据对预测模型进行训练和预测。最后根据预测评价指标(预测精度、最大相对误差、相对均方误差),对结果分析评价。同时与未降维处理的多元WNN预测模型、单元WNN预测模型进行比较。实验结果表明:基于多元相空间重构的数控机床运动精度预测模型的预测精度高,且经降维处理的多元预测模型所得相对均方误差比其他两个模型低了一个数量级。说明了本文提出的预测模型能够有效跟踪数控机床运动精度变化规律。
[Abstract]:Precision is an important performance index of NC machine tools. The accuracy of NC machine tools directly affects the precision and quality of machining parts, and reflects the technical level and competitiveness of the whole manufacturing industry. China is a large consumer of CNC machine tools, but not a powerful country in NC machine tool manufacturing. The demand and manufacturing force are in an unbalanced state, especially for high-grade NC machine tools. Compared with imported CNC machine tools, the precision retention of domestic NC machine tools is not high, so it is an urgent problem to improve the precision retention of domestic CNC machine tools. In this paper, the motion accuracy of NC machine tools is analyzed and forecasted. According to the results of analysis and prediction, measures can be taken to reduce or eliminate the error sources before the motion accuracy of NC machine tools fails, so as to improve the machining accuracy of products. The service time of NC machine tool is prolonged and the accuracy retention is improved. In addition, it can also guide the maintenance of machine tools to prevent economic losses caused by disrepair. In this paper, the motion precision phase space is established based on the multivariate phase space technique, and the low dimensional time series is mapped to the high dimensional phase space to restore the chaotic attractor. The orderly state of being exposed to a complex representation. Principal component analysis (PCA) is introduced to reduce the dimension of the reconstructed phase space and simplify the model structure. Then the prediction model of wavelet neural network is constructed. The phase coordinate of motion accuracy is taken as input and the variation of roundness error is output. The prediction model is trained and studied so as to realize the prediction of motion accuracy of NC machine tools. The main research work of this paper is as follows: 1) build an experimental platform to test the motion accuracy of CNC machine tools and obtain time series data of multiple motion accuracy characteristic quantities. The maximum Lyapunov exponent is used to analyze the chaotic characteristics of the sequence of motion accuracy of NC machine tools, and the phase space of motion accuracy of NC machine tools is reconstructed. C-C algorithm is used to calculate the phase space reconstruction parameters of each time series of motion accuracy characteristic variables. Then the redundant information is filtered by principal component analysis to simplify the phase space structure. The dimension of phase space of motion precision is used as the number of neurons in the input layer of wavelet neural network, which avoids the trial and error method, and ensures the completeness of input information. The wavelet neural network prediction model of motion accuracy of NC machine tool is constructed. The Morlet wavelet is chosen as the neuron excitation function of hidden layer, the wavelet neural network structure is analyzed and determined, the parameter correction algorithm of the prediction model is given, and then the prediction model is trained and predicted by the reconstructed motion precision phase space data. Finally, according to the prediction evaluation index (prediction accuracy, maximum relative error, relative mean square error), the results are analyzed and evaluated. At the same time, it is compared with the multivariate WNN prediction model and the unit WNN prediction model without dimensionality reduction. The experimental results show that the prediction accuracy of the motion accuracy prediction model of NC machine tools based on multi-phase space reconstruction is high, and the relative mean square error of the multivariate prediction model treated by dimensionality reduction is one order of magnitude lower than that of the other two models. The prediction model presented in this paper can effectively track the movement accuracy of NC machine tools.
【学位授予单位】：重庆理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG659

【参考文献】