摘要
由于铣削加工中发生颤振会极大地降低工件的加工质量,铣削振动状态的高效与精准辨识一直是颤振研究的热点问题之一。基于LetNet-5经典卷积网络提出一维卷积网络模型,直接对时域铣削力信号进行处理与识别,针对信号量较少与数据不均衡等问题,采用重叠-随机协同采样的方法对数据进行处理。应用T-分布随机邻域嵌入技术可视化模型在训练集上的学习进程并对端到端的学习目标进行验证。对比基于支持向量机与卷积神经网络识别策略,所提方案在测试集上取得了最高的96.17%准确率,识别结果表明:该方法相较于对比方法过程简单、识别快速且辨识准确率高。
Since the occurrence of chatter in milling may greatly reduce the machining quality of workpieces,the efficient and accurate identification of milling vibration state has always been one of the hot issues in chatter research.Based on the LetNet-5 classical convolution network,a one-dimensional convolution network model is proposed,which can directly process and identify milling force signals in the time domain.To solve the problems such as less semaphore and unbalanced data,the overlapping-random collaborative sampling method is adopted to process data.The T-distribution random neighborhood embedding technology is applied to visualize the learning process of the training set model and verify the end-to-end learning objectives.Comparing the recognition strategy of the standard support vector machine(SVM)with that of the convolution neural network,the proposed model achieves the highest accuracy of 96.17%on the test set.The recognition results show that the proposed model is simple,fast and accurate.
作者
郑华林
张冲
何勇
ZHENG Hualin;ZHANG Chong;HE Yong(School of Mechanical and Electrical Engineering,Southwest Petroleum University,Chengdu 610065,China;China National Petroleum Corporation Chuanqing Drilling Training Center,Chengdu 610213,China)
出处
《机械科学与技术》
CSCD
北大核心
2023年第7期1081-1087,共7页
Mechanical Science and Technology for Aerospace Engineering
基金
四川省科技厅重点研发项目(19ZDZX0055)。
关键词
铣削振动
状态识别
一维卷积神经网络
milling vibration
state recognition
one-dimensional convolution neural network model
