【作者】 杨超；

【导师】 李宏坤； 李特；

【作者基本信息】 大连理工大学 ， 机械工程（专业学位）， 2023， 硕士

【摘要】 切削加工中,刀具是金属工件成形的重要执行元件,其锋利程度直接影响工件质量、加工效率、生产成本以及生产稳定性。当刀具产生严重磨损而没有及时更换刀具时,会导致刀具断裂甚至发生工件报废的情况。因此,对刀具磨损状态的识别在实际生产中具有重要价值。本文通过监测加工过程中数控机床进给电机的电流信号,利用深度学习算法进行特征提取并识别刀具的磨损状态。论文的主要研究内容如下:(1)对国内外刀具磨损状态监测研究现状进行论述,其中介绍常用的监测信号、特征提取方法和监测模型,对现有的监测方法进行综合分析。研究刀具的磨损机理,对进给驱动模型进行分析,理论上论证了使用数控机床进给电流作为监测信号的可行性。设计进给电流和多信号采集方案,利用不同的数据预处理方式对两种方案的数据进行样本制作。根据刀具磨损量划分刀具磨损状态,并完成数据样本的划分。(2)针对原有变分自编码器识别效果差的问题,构造了深度约束变分自编码器。增加一层编码网络以挖掘样本数据的深层特征信息,从而增强模型的提取能力;利用约束条件控制采样层中隐藏变量的变化范围,使刀具在不同磨损状态下的特征分布更加集中,以增强模型的稳定性。(3)通过对比分析进给电流信号和铣削力信号,验证两者之间的联系,进而表明进给电流信号的实际可行性。利用深度约束变分自编码器和极限学习机识别铣刀的磨损状态,将完成预处理的进给电流信号输入到深度约束变分自编码器中,通过无监督方式提取样本数据中的特征信息;将提取到的数据特征输入到极限学习机中,通过有监督方式分类识别特征信息,从而识别铣刀的磨损状态。(4)采用不同直径刀具的进给电流数据集,验证网络模型的有效性;采用相同直径刀具的进给电流和多信号融合数据集,验证进给电流信号的优越性。利用Lab VIEW软件设计并开发刀具磨损状态监测系统,采集并分析进给电流信号,获取刀具磨损等级。更多还原

【Abstract】 In the cutting process,the tool is an important executive element of metal workpiece forming,its sharpness directly affects the quality of the workpiece,processing efficiency,production costs and production stability.When the tool become severely worn and the tool is not replaced in time,it can lead to the tool breakage and even workpiece scrap.Therefore,the identification of tool wear state is of great value in practical production.In this paper,by monitoring the current signal of the feed motor of CNC machine tool during the machining process,the deep learning algorithm is used to extract feature and identify the wear state of the tool.The main research contents of this paper are as follows:(1)Discusses the research of tool wear state monitoring at home and abroad,which introduces commonly used monitoring signals,feature extraction methods and monitoring models,and the existing monitoring methods are analyzed comprehensively.The wear mechanism of the tool is studied,the feed driving model is analyzed,and the feasibility of using the feed current of the CNC machine tool as monitoring signal is demonstrated theoretically.The feed current and multi-signal acquisition schemes are designed,and the data samples of the two schemes are made by using different data preprocessing methods.The tool wear state is divided according to the tool wear amount,and the data sample is divided.(2)In order to solve the problem of poor recognition effect of the original variational autoencoder,a deep-constrained variational auto-encoder is constructed.A layer of coding network is added to mine the deep feature information of the sample data,so as to enhance the extraction ability of the model.Constraint condition is used to control the variation range of hidden variables in the sampling layer,so that the feature distribution of the tool under different wear states is more concentrated,so as to enhance the stability of the model.(3)By comparing and analyzing the feed current signal and the milling force signal,verify the connection between them,and then show the practical feasibility of feed current signal.The wear state of milling tool is identified by using the deep-constrained variational auto-encoder and the extreme learning machine.And the pre-processed feed current signal is input into the deep-constrained variational auto-encoder,the feature information of the sample data is extracted by unsupervised manner.The extracted data features are input into the extreme learning machine,and the feature information is classified and recognized by supervised manner,so as to identify the wear state of the milling tool.(4)The feed current data sets of the different diameter tools are used to verify the validity of the network model.The feed current and multi-signal fusion data sets of the same diameter tool are used to verify the superiority of the feed current signal.The tool wear state monitoring system is designed and developed by using Lab VIEW software,and the feed current signal is collected and analyzed to obtain the tool wear level.更多还原