基于GMM的纳米制造刀具磨损状态在线识别  

On-line diagnosis of tip-wear in nano-machining based on Gaussian mixture model

作　　者：程菲[1,2] 江子湛 CHENG Fei;JIANG Zizhan(School of Management,Hangzhou Dianzi University,Hangzhou 310018,China;School of Big Data and Artificial Intelligence,Anhui Institute of Information Technology,Wuhu 241000,China)

机构地区：[1]杭州电子科技大学管理学院,浙江杭州310018 [2]安徽信息工程学院大数据与人工智能学院,安徽芜湖241000

出　　处：《计算机集成制造系统》2024年第11期4075-4086,共12页Computer Integrated Manufacturing Systems

基　　金：浙江省社会科学界联合会资助项目(2022N31);安徽省自然科学基金资助项目(2008085MF201);安徽省哲学社会科学规划一般项目(AHSKY2021D142)。

摘　　要：为满足纳米制造刀具磨损状态在线诊断对时间和精度的要求,采用跨物理数据融合建模方案,建立具有物理一致性的高斯混合模型(GMM),以动态识别原子力显微镜(AFM)尖端状态。随机抽取历史加工数据,提取特征参数并进行训练,获得3维GMM模型并预存;以加窗分帧的形式,截取连续过程中短时段纳米加工力时变信号,构成瞬时稳态数据空间;以尖端旋转周期为时间单位,计算横向加工力的特征参数:极大值、峰-峰值和方差;采用马氏距离检测并去除异常值。使用预存的GMM模型,对每帧特征参数聚类,识别尖端磨损状态;根据连续分析帧的尖端失效点数据变化曲线,探测跟踪尖端状态。实验证明该算法平均识别精度为0.8917,平均召回率为0.963;每2000个点的最长识别时间为31ms,平均识别时间为23.97ms,适用于大规模纳米制造的刀具磨损在线自动诊断。To meet the requirements of time and accuracy for online diagnosis of tool wear state in nano-manufacturing,a Cross-Physical Data Fusion(CPDF)scheme was adopted to establish a physically-consistent Gaussian Mixture Model(GMM)to dynamically identify the tip-wear of Atomic Force Microscope(AFM).Historical processing data were randomly selected and feature parameters were extracted and trained to acquire the 3D GMM model and then pre-stored.Through the windowing and framing,the time-varying signals of nano-machining force in a short period of time in the continuous process were intercepted to form an instantaneous steady-state data space.Took the tip rotation period as the time unit,the feature parameters of the transverse machining force were calculated,which included the maximum value,peak to peak value and variance.Outliers were detected and removed using Mahalanobis Distance.The pre-stored GMM model was used to cluster the feature parameters in each frame to identify the tip wear state,and the tip state was detected and tracked based on the change curve of tip failure points data in continuous analysis frames.Experiments showed that the average recognition accuracy of the algorithm was 0.8917 and the average recall was 0.963.The longest recognition time per 2000 points was 31MS,and the average recognition time was 23.97ms.All of these findings indicate that GMM was suitable for online automatic diagnosis of tool wear in large-scale nano-manufacturing.

关 键 词：纳米加工 刀具磨损在线诊断 高斯混合模型 机器学习 数据融合集成制造 

分 类 号：TG71[金属学及工艺—刀具与模具]