基于集成正交实验与CNN方法的微细电火花微孔加工精度  

Micro-electrical Discharge Machining of Micro-holes Based on Integrated Orthogonal Experiments and CNN Methods

作　　者：莫远东 王雅芝 黄舒琦 钟佳俊 MO Yuandong;WANG Yazhi;HUANG Shuqi;ZHONG Jiajun(School of Electromechanical Eng.,Lingnan Normal Univ.,Zhanjiang 524048,China)

机构地区：[1]岭南师范学院机电工程学院,广东湛江524048

出　　处：《工程科学与技术》2024年第4期261-272,共12页Advanced Engineering Sciences

基　　金：广东省普通高校青年创新人才类项目(2022KQNCX041);岭南师范学院自然科学一般项目(YB2111)。

摘　　要：针对改善微细电火花微孔加工精度问题,提出一种集成正交试验与卷积神经网络的方法。首先,利用正交试验研究进给速度、主轴转速、脉冲占空比和脉冲频率4个因素对微细电火花加工H62黄铜微孔入口过切(En-OV)、出口过切(ExOV)和锥度(TA)的影响规律及其最优加工工艺参数。然后,采用基于PyTorch框架的卷积神经网络对试验结果进行预测。结果表明:对于微孔加工入口过切,各因素按影响程度由高到低依次为脉冲占空比、脉冲频率、进给速度和主轴转速;对于微孔加工出口过切,各因素按影响程度由高到低依次为进给速度、主轴转速、脉冲占空比和脉冲频率;对于微孔加工锥度,各因素按影响程度由高到低依次为进给速度、脉冲占空比、主轴转速和脉冲频率。综合考虑并分析各因素之间关系,通过验证实验得到微孔加工精度最优组合参数:进给速度为0.02mm/s、主轴转速为1000r/min、脉冲占空比为60%、脉冲频率为3000Hz。微孔加工入口过切、出口过切和锥度的预测值与试验真实值吻合程度较高,两者相对误差均小于12%。基于PyTorch框架的卷积神经网络(CNN)具有较高的预测准确性,满足实际生产加工需求,为微细电火花微孔加工精度预测研究提供了新方法,也为实际生产加工指导提供了依据。Objective To improve the precision of the micro-electrical discharge machining(micro-EDM)of micro-holes,a pioneering approach is undertaken in this study.Deep learning is applied innovatively to micro-hole micro-EDM,weaving together orthogonal experiments and deep learning to present an intricate methodology,i.e.,an integrated fusion of orthogonal experiments and convolutional neural networks(CNNs).This synthesis strives to secure a robust dataset for subsequent predictive analyses of experimental outcomes via a CNN while concurrently minimizing the number of experiments.Methods The initial phase involves a meticulous L27(313)orthogonal experiment,featuring four factors and three levels.The intricate impact patterns and optimal processing parameters for feed rate,spindle speed,pulse duty cycle,and pulse frequency regarding entrance overcut(EnOV),exit overcut(ExOV),and taper angle(TA)in the micro-EDM of H62 brass micro-holes are investigated meticulously in this study.Rigorous range and variance analyses are conducted on the experimental results,complemented by the deployment of scanning electron microscopy(SEM)to scrutinize the machined morphology for result validation.Subsequently,to validate the precision and applicability of the model predictions,81 groups of data from orthogonal experiments on entrance overcut,exit overcut,and taper angle serve as predictive data for a CNN based on the Py-Torch framework.Eleven groups of data are chosen meticulously as prediction samples,leaving the remaining 70 groups for training samples to predict experimental results.Results and Discussions The results show that feed rate,spindle speed,pulse duty cycle,and pulse frequency wield significant influence over entrance overcut,exit overcut,and taper angle in the micro-EDM of H62 brass micro-holes.For the entrance overcut in micro-hole machining,the hierarchy of impact intensity from high to low is pulse duty cycle,pulse frequency,feed rate,and spindle speed,and the identified optimal combination of parameters is a feed rate of

关 键 词：微细电火花加工 正交实验 卷积神经网络 H62黄铜 微孔 

分 类 号：TG661[金属学及工艺—金属切削加工及机床]