基于多维时间序列的数控机床状态预测方法研究

中文摘要: 随着数控机床结构复杂化以及运行状态数据呈现多样性、时序性的特点，为了有效解决数控机床未来状态难以准确预测的难题，提出一种基于多维时间序列的数控机床状态预测方法。首先，采用OPC（OLE for process control）技术进行数控机床状态数据采集，结合Min-max标准化和自回归移动平均模型完成了数据预处理，建立了多维时间序列状态模型及度量模型，采用特征向量、特征趋势距离标示状态模型，并利用差异度进行多维时间序列状态匹配分析。其次，通过建立时间窗口滑动模型，利用时间窗口长度和滑动时长获取数控机床历史状态集合，进一步提出基于窗口滑动的多重匹配技术，利用 β-耦合相似度量标准寻找与当前状态矩阵相似度最大的历史状态集合，并根据相似性阈值得到最优滑动时长和预测时长。然后，采用密度空间聚类算法进行状态序列分析，得到了表征机床当前时刻状态的最佳历史状态矩阵，并以此状态的下一时刻作为预测状态。最后，对数控机床主轴四项参数开展了数控机床状态预测实验，通过状态序列相似性分析得到最佳预测时长为24 s，滑动单位为2 s，并利用状态序列聚类分析完成状态序列匹配。预测结果表明，基于多维时间序列的状态预测方法的最大误差、平均误差、均方误差和相对误差均低于传统的AR预测模型，验证了所提出的状态预测方法的有效性和准确性。
Abstract:With the structural complexity of machine tool and the diversity and time-sequence characteristic of state data,in order to solve the problem that the future state of machine tool is difficult to accurately predict,a novel state prediction method based on multidimensional time series was proposed.Firstly,the ole for process control (OPC) technology was used to collect data of machine tool,and the Min-max normalization and autoregressive moving average model were adopted to data preprocessing.The state and evaluation models of multidimensional time series were established.Meanwhile,the feature vector and trend distance were also used to represent state model,and then the state match of multidimensional time series was completed by difference degree analysis.Secondly,through constructing the time sliding window technique,the historical state sets of machine tool were obtained by the length of time window length and sliding.On this basis,multiple matching technique based on window sliding was developed,and then the β-coupling similarity metrics was also used to find a set of historical states that were the most similar to the current state matrix.According to the similarity threshold,the optimal sliding time and prediction time were obtained.Further,the density-based spatial clustering algorithm was adopted to perform state series analysis,and the optimal historical sate matrix which can represent the current state of machine tool was obtained,and then the next state was regarded as prediction state.Finally,the state prediction experiments were carried out for four parameters of machine tool spindle.The optimal prediction time and sliding unit were 24 s and 2 s respectively,and then the state-sequence matching was completed by using state-sequence clustering analysis.The prediction results showed that the maximum error,mean error,mean square error and relative error of the matrix and vector state prediction approach based on multidimensional time series were lower than those of