摘要
半导体生产过程是典型的间歇过程,针对其过程数据的多模态、多阶段、模态结构不同和批次不等长等特点,提出了基于统计模量的局部近邻标准化和K近邻相结合的故障检测方法(SP-LNS-KNN)。首先计算样本的统计模量,其次对样本的统计模量使用其局部K近邻集进行标准化,最后计算样本与其前K近邻距离,得到平均累积距离D作为检测指标,进而对工业过程进行在线故障检测。统计模量保留了数据的主要信息,将二维样本数据简化为一维数据;局部近邻标准化可以有效降低中心漂移、模态结构差异明显的影响。SP-LNS-KNN不仅能够对大故障实现检测,并且能够提高对小模态的微弱故障的检测能力。使用SP-LNS-KNN对一个实际半导体生产过程数据进行故障检测实验,并将实验结果与PCA、KPCA、LOF和FD-KNN方法的结果进行对比分析,验证了方法的有效性。
A semiconductor manufacturing process is operating a typical batch process.According to the characteristics of multi-mode,multi-stage,different modal structure and unequal batch length of semiconductor process data,this paper proposed a fault detection method that the combination of the local neighborhood statistical standard modulus and K-nearest neighbor based on statistics pattern(SP-LNS-KNN).First,it calculated statistical patterns of modulus samples,secondly standardized each sample using the local K-neighbor.Then it calculated the distances between each sample and its K-nearest neighbor,got average cumulative distance D and counted D as the testing index.Furthermore,the industrial process were detected faults on-line.The statistics pattern retained the primary information of the data and simplified the two-dimensional sample data into one-dimensional data.The local neighborhood normalization could effectively reduce the influence of the center shift and the va-riance of the modal structure obviously.SP-LNS-KNN could not only detect large faults,but also improve the ability to detect small faults in small modes.At last this paper conducted a practical fault detection experiment on a semiconductor manufactu-ring process using SP-LNS-KNN and compared the result with the results of the PCA,KPCA,LOF and FD-KNN methods to verify the effectiveness of the method.
作者
冯立伟
张成
谢彦红
李元
逄玉俊
Feng Liwei;Zhang Cheng;Xie Yanhong;Li Yuan;Pang Yujun(Dept.of Science,Shenyang University of Chemical Technology,Shen-yang 110142,China;Research Center for Technical Process Fault Diagnosis&Safety,Shenyang University of Chemical Technology,Shen-yang 110142,China)
出处
《计算机应用研究》
CSCD
北大核心
2018年第11期3307-3310,3314,共5页
Application Research of Computers
基金
国家自然科学基金重点资助项目(61490701
61673279)
辽宁省教育厅基金资助一般项目(L2015432)
辽宁省自然科学基金资助项目(2015020164)
关键词
标准化
K近邻
多模态
故障检测
统计模量分析
standardization
K-nearest neighbor
multimode
fault detection
statistics pattern analysis
作者简介
冯立伟(1980-),男,辽宁沈阳人,讲师,硕士,主要研究方向为过程故障检测(feng-li-wei@163.com);张成(1979-),男,副教授,硕士,主要研究方向为过程故障诊断分析;谢彦红(1964-),女,辽宁沈阳人,教授,博士,主要研究方向为过程控制与故障诊断;李元(1964-),女(通信作者),教授,博士,主要研究方向为基于数据驱动复杂过程故障诊断;逄玉俊(1954-),男,辽宁东港人,教授,硕士,主要研究方向为模式识别、智能系统、计算机及应用、控制理论与工程、智能信息处理等.
