摘要 为了提高预测机械加工表面粗糙度的精度,提出了基于Copula分布估计算法(estimation of distribution algorithm,EDA)优化BP神经网络的方法.以铣削45#钢为试验对象,采用控制变量法进行切削试验.在线测量主切削力、轴向力、径向力和振幅,并进行数据处理,得到相应切削力的平均值、标准差、均方根值及振幅,同时离线测量二维粗糙度Ra、三维粗糙度平均值Sa和均方根值Sq.对切削分力的平均值、标准差、均方根值及振幅与粗糙度做相关性分析,选择Kendall秩相关系数最大的主切削力平均值作为输入变量,输入BP神经网络和基于Copula EDA优化BP神经网络,进行训练和预测.试验结果表明:基于Copula EDA优化BP神经网络的预测精度总体高于BP神经网络的预测精度,对Ra,Sa和Sq的平均预测精度分别达到91.98％,91.03％和89.10％.
Abstract： To improve the prediction accuracy of surface roughness of machined surface, the method of optimizing BP neural network was proposed by Copula estimation of distribution algorithm (EDA). The experiments of milling 45＃ steel were conducted by the control variable method. The main cutting force, axial force, radial force and vibration amplitude were measured on line, and the corresponding average value, standard deviation, RMS values of cutting force and vibration amplitude were obtained. The twodimensional surface roughness Ra, threedimensional roughness average Sa and RMS Sq were measured offline. The correlation analysis was carried out among cutting force component mean value, standard deviation, root mean square value, vibration amplitude and roughness. The average value of main cutting force with the largest Kendall rank correlation coefficient was selected as input variable. The average main cutting force was input into BP neural network and another BP neural network optimized by Copula EDA for training and predicting. The experimental results show that the prediction accuracies of BP neural network based on Copula EDA is higher than that of BP neural network, and the average prediction accuracy of Ra, Sa and Sq are 91.98％, 91.03％ and 89.10％, respectively.
