基于多传感器信号的主轴回转误差在线回归预测方法研究OA北大核心CSTPCD
Study on Regression Prediction Method of Multi Signal Spindle PSO-SVM Rotation Error Based on LMD
通过对主轴回转误差形成机理分析,基于多传感器信号构建了主轴回转误差回归预测模型.首先,使用LMD分解方法和皮尔逊相关系数算法对主轴回转误差密切相关的主轴前轴承振动信号、主轴电流信号和前轴承声发射信号进行特征量提取和寻优降维,克服了以往机床主轴回转误差预测中原始信号类型过于单一的问题.其次,针对各监测信号与机床主轴回转误差之间的非线性问题,利用支持向量机模型特有的RBF核函数实现多输入量下的非线性预测、找到数据之间的复杂关系,但模型中宽度系数σ、惩罚因子C和不敏感损失系数ε的有效确定是建立其RBF核函数的难题,为此建立了基于粒子群算法的支持向量机模型对主轴回转误差进行回归预测.再次,为评价该模型的有效性,提出了基于均方误差、平均绝对误差和决定系数的主轴回转误差回归预测模型评价方法.最后,使用i5m4数控加工中心对上述预测模型进行了实验研究,结果表明:该PSO-SVM回归预测模型的均方误差为0.19%,平均绝对误差为4.58%,决定系数为0.923 7,相对于优化前模型,该PSO-SVM回归预测模型可准确有效实现主轴回转误差的预测.
By analyzing the formation mechanism of spindle rotation errors,regression forecasting model of spindle rotation errors which based on multi-sensor signal was built.Firstly,the LMD method and Pearson correlation coefficient method were used to extract feature values and optimize dimensionality reduction for the vibration signals,current signals,and acoustic emission signals of the front bearing,which solve the problem that types of original signals used in prediction of spindle rotation error of CNC machine tools were too single.Secondly,aiming at the nonlinear problems between various monitoring signals and rotation errors of the machine tool spindle,the RBF kernel function was used to achieve nonlinear prediction under multiple inputs and find complex relationships between datas.However,to establish RBF kernel function,the effective determination of width coefficient σ,penalty factor C and insensitive loss coefficient ε was a challenge in the model.Therefore,a support vector machine model based on particle swarm optimization algorithm was established to predict the spindle rotation errors.Once again,to evaluate the effectiveness of the model,a regression prediction model evaluation method for spindle rotation error based on mean square error,mean absolute error and coefficient of determination was proposed.Finally,experimental research was conducted on above prediction model in the i5m4 CNC machining center.The results showed that the mean square error of PSO-SVM regression prediction model was 0.19%,the average absolute error was 4.58%,the coefficient of determination was 0.923 7.Compared with the model before optimization,the PSO-SVM regression forecasting model can predict the spindle rotation errors accurately and effectively.
迟玉伦;宋卓阳;王国强;姚磊
上海理工大学机械工程学院,上海 200093
几何量计量主轴回转误差多传感器信号PSO-SVMLMD皮尔森相关系数
geometric measurementspindle rotation errormulti signalPSO-SVMLMDPCA
《计量学报》 2024 (009)
1300-1313 / 14
国家自然科学基金(51605294)
评论