【作者】 王强；

【导师】 郝小忠； 崔海涛；

【作者基本信息】 南京航空航天大学 ， 数字化设计与制造， 2019， 硕士

【摘要】 数控加工过程中刀具状态异常会导致零件质量不合格甚至报废,而不合理的换刀会造成成本浪费。预测刀具寿命可以有效避免因刀具状态异常导致的零件质量问题,提高刀具的使用率。但刀具磨损过程复杂多变,刀具剩余寿命受工况影响难以准确预测。针对以上问题,本文对数控加工刀具寿命预测问题进行了深入研究。论文主要工作如下:(1)针对刀具监测数据维度高难以有效处理的问题,研究了基于时频域分析和受限玻尔兹曼机的刀具磨损信号特征提取方法,充分挖掘了监测数据中刀具磨损相关信息,并采用敏感特征筛选和特征矩阵降维的方法降低刀具磨损相关信号特征的维度,实现了信号特征的有效提取。(2)针对刀具实际磨损量难以快速准确获取的问题,提出了一种基于机器视觉的刀具磨损量在线自动测量方法,在自动获取刀具磨损图像的基础上,通过刀具特征线准确定位磨损关键区域,利用滑窗遍历法重建刀具原始边界,完成刀具磨损区域提取后自动计算刀具磨损量,提高了铣刀磨损量直接测量的精度与效率。(3)针对数控加工中刀具剩余寿命难以准确预测的问题,提出了一种基于在线学习的刀具寿命动态预测方法,该方法以改进的长短时记忆深度学习网络模型为基础,建立了刀具寿命预测模型,以刀具磨损量和剩余寿命为样本标签进行模型训练,融入在线学习模块,以在线测量的数据在线更新模型,实现了变工况下刀具寿命的预测。(4)基于以上研究,在LabVIEW平台上开发了刀具寿命动态评估系统,并设计实验对本文所提方法进行验证。更多还原

【Abstract】 The abnormal status of cutting tools in NC machining will lead to unqualified or even scrapped parts,and unreasonable tool replacement will lead to waste of cost.The prediction of tool life can effectively avoid part quality problems caused by abnormal tool status and improve the tool utilization rate.However,the tool wear process is complicated,and the residual life of tool is difficult to be predicted accurately by the influence of working conditions.For the above questions,the problem of tool life prediction in NC machining is deeply studied in this thesis.The main work of this thesis is as follows:(1)Aiming at the problem that the high dimension of monitoring data is difficult to handle effectively,the feature extraction method of tool wear signals based on the time-frequency domain analysis and RBM is studied,and the related information in monitoring data is fully excavated.The dimension of tool wear related signals is effectively reduced by selection of sensitive feature and reduction by characteristic matrix,and the effective feature extraction of the signals is realized.(2)To solve the problem that the tool wear is difficult to obtain quickly and accurately,an online automatic measurement method of tool wear based on machine vision is proposed.The region of interest is located by tool feature lines accurately,and the original edge of tool is reconstructed by traversing the sliding window.The tool wear is calculated automatically after the tool wear area is extracted,which improves the accuracy and efficiency of direct measurement of milling wear.(3)In order to solve the problem that tool residual life is difficult to predict accurately in NC machining,a dynamic method of tool life prediction based on online learning is proposed.Based on the modified long-short term memory network model,the tool life prediction model is established.Model training is carried out with tool wear and residual life as sample labels,and the online learning module is integrated for the online update of model based on the measured data,thus realizing the prediction of tool life under variable working conditions.(4)Based on the above research,the tool life dynamic evaluation system was developed on LabVIEW,and experiments were designed to verify the method proposed in this thesis.更多还原