工业大数据驱动的高维过程质量稳健监控模型的构建与优化  

Construction and Optimization of Robust Quality Monitoring Model in High-Dimensional Process Motivated by Industrial Big Data

作　　者：张帅 杨剑锋[2] 闫莉[1] ZHANG Shuai;YANG Jianfeng;YAN Li(Management Engineering School,Henan University of Engineering,Zhengzhou Henan 451191,China;Busniess School,Zhengzhou University,Zhengzhou Henan 450001,China)

机构地区：[1]河南工程学院管理工程学院,河南郑州451191 [2]郑州大学商学院,河南郑州450001

出　　处：《机床与液压》2024年第18期48-53,共6页Machine Tool & Hydraulics

基　　金：国家自然科学基金联合基金项目(U1904211);国家自然科学基金面上项目(71672182);河南省重点研发专项(241111210200;241111222700);河南省科技攻关项目(232102211040)。

摘　　要：由于工业大数据存在变量维度高、价值密度低、存在离群点等因素,监控模型难以准确挖掘海量数据中关键波动信息,容易产生较高的误警率,影响产品生产质量。为解决这个问题,提出一种基于最小行列式法和变量选择算法的高维过程稳健监控模型。运用最小协方差行列式(MCD)方法估计稳健的均值向量和协方差矩阵;构建似然比检验统计量,通过增加惩罚项得到变量选择优化函数;结合MCD和变量选择得到稳健的监控统计量,利用Monte Carlo方法得到监控用控制限;最后,通过仿真数据和薄膜沉积过程实际数据对所提方法进行实证研究。结果表明:所提方法相比Hotelling T2和VS控制图具有较高的异常识别精度和鲁棒性,在存在离群点的高维过程质量监控中提高了对异常波动识别的稳健性,达到了期望的监控效率。In the environment of industrial big data,due to some factors including high-dimensional,low value density and outlier points,it is difficult for the monitoring model to dig the key fluctuation information from mass data,which will lead to high false alarm rate and affect the product quality.To overcome this problem,a novel robust monitoring model in the high-dimensional process was proposed based on minimum covariance determinant estimation and variable selection algorithm.The minimum covariance determinant estimation(MCD)method was applied to estimate the robust mean vector and covariance matrix;the likelihood ratio test statistic was constructed,the variable selection optimization function was obtained by adding a penalty term;robust monitoring statistics was obtained combining MCD and variable selection,the control limit for monitoring was obtained using Monte Carlo method;finally,the proposed method was empirically studied by simulation data and actual data of the film deposition process.The results show that compared with Hotelling T2 and VS control charts,the proposed method has high abnormal identification accuracy and robustness,the robustness of abnormal fluctuations identification is improved in high-dimensional process quality monitoring in the presence of outlier points,and the desired monitoring efficiency is achieved.

关 键 词：工业大数据 最小协方差行列式估计 过程质量监控 变量选择算法 

分 类 号：TH166[机械工程—机械制造及自动化] TG506[金属学及工艺—金属切削加工及机床]