基于优化的组合协方差高斯过程的表面粗糙度预测

Prediction of Surface Roughness Based on Optimized Combined Covariance Gaussian Process Regression

加工过程中产生的表面质量(如粗糙度)的数据序列包含多种特征,为能捕捉更多的数据特征,提高表面粗糙度的预测精度,提出采用组合协方差的高斯过程回归(CGPR)模型预测表面粗糙度,来捕捉数据特征中的线性特征和非线性特征;为获得CGPR模型的最佳超参数组合,采用人工蜂群(ABC)优化算法对超参数寻优,形成人工蜂群—组合协方差的高斯过程回归(ABC-CGPR)模型。通过45钢的车削试验,基于不同切削用量和刀具结构,建立了各类不同组合协方差的ABC-CGPR预测模型和单一协方差的ABC-GPR预测模型,并对比其预测性能,结果展示CGPR预测模型相比单一的GPR预测模型具有更高的预测精度,其中线性协方差函数与Matern协方差函数组合的预测精度最高,为实际加工中选取满意的预测模型提供了有效的指导。

The data sequence of surface quality (such as roughness) produced in the process of machining contains many features.In order to capture these data features and improve the prediction accuracy of surface roughness,this paper proposed a combined covariance gaussian process regression (CGPR) model to predict surface roughness,and to simultaneously capture the linear and non-linear features of data features.In order to obtain the optimal combination of hyperparameters,the artificial bee colony (ABC) optimization algorithm was used to optimize the hyperparameters,and the artificial bee colony-combined covariance gaussian process regression (ABC-CGPR) model was formed.Based on different cutting parameters and tool structure,various ABC-CGPR prediction models with different combination covariance functions and ABC-GPR prediction models with single covariance function were established through turning experiments of 45 steel,and their prediction performance was compared.The results show that the combined covariance prediction models have higher prediction accuracy than the prediction models with single covariance function,and the model that combined by the linear covariance function and the Matern covariance function has the highest prediction accuracy,which provides effective guidance for selecting satisfactory prediction models in actual processing.