Abstract: To accurately and quickly predict the energy consumption of turning and drilling processes,a data-driven methodology to predict energy consumption of machining a part was proposed,which included four key technologies that were manufacturing data acquisition and preprocessing,preprocessing of feature attribute,algorithm for feature selection and energy consumption prediction.The feature selection was achieved by combining sample classification and RReliefF algorithm.The energy consumption was predicted using three algorithms that were neural network,support vector regression and random forest,and the prediction accuracy was improved by adjusting the parameters of the algorithms.Experiments were conducted to validate the proposed method.The energy consumption of cylindrical turning and drilling processes of parts was predicted using the proposed methodology and compared with the measured energy.Results showed that the proposed method could be used to identify the main factors influencing the machining energy consumption.The average prediction errors range from 4.94% to 9.94% and decreased as the number of training samples increasing.The neural network algorithm could achieve the lowest prediction error,which was lower than those obtained using existing method.The proposed methodology had a big potential for industrial applications.

Key words: data-driven, turning, drilling, feature selection, machining, energy consumption prediction

摘要： 为准确快捷地预测零件车削和钻削加工工艺过程的电能消耗,提出一种基于数据驱动的能耗预测方法,包括能耗数据采集和预处理、特征属性预处理、特征选择算法和能耗预测算法4个关键技术。将样本分类和RReliefF算法结合进行特征选择,采用神经网络、支持向量回归、随机森林3种算法预测能耗,并通过对算法进行调参提高预测精度。在此基础上进行了实验研究,应用所提方法预测零件外圆车削和钻削加工能耗,并同实测能耗进行比较。案例分析结果表明,所提方法能够分析得出零件加工能耗的主要影响因素,3种算法的平均预测误差在494%~994%之间,误差随着训练样本的增加逐步下降,其中神经网络算法的预测误差最小,低于现有方法,具有很大的应用潜力。

关键词: 数据驱动, 车削, 钻削, 特征选择, 机械加工, 能耗预测