数据驱动的多元混杂数据过程控制与应用研究

Under the consideration of current problems of advanced manufacturing systems, the project titled by data-driven process control and application for multivariate mixed data is proposed to deal with the process data with the characteristics of the variety of data resources, a huge number of data, mixed variable types. Three issues are provided including data-driven control charts pattern recognition (CCPR), data-driven process control for mixed variable types, and data-driven application of process monitoring. Although the current research results correlated with these issues are limited, it will take advantages of data-driven methods’ ability in addressing the baseline operation of a process or a mention of a Phase I analysis, and studying process data comprising of a variety of data types. It will be deployed according to the structure from univariate variable to multivariate variables, from low-dimensional to high-dimensional problem, from methodology to application; the roadmap from statistical analysis, to the study of methods and tools of process monitoring, and then to CCPR of mixed data types; the roadmap from data collection and matching, to the use of methods and tools for process monitoring, and then to the description of analytical results. We will cooperate with HEXAGON Metrology (Qingdao) Company Limited, do research on application of methods and tools for process monitoring at HEXAGON’s digital workshop, and then provide solution of data-driven process monitoring for enterprises.

考虑到国内外先进制造系统中质量控制遇到的新问题，面对过程数据所具有的来源多样化、数据量激增、类型混杂的特点，本项目提出了数据驱动的多元混杂数据过程控制与应用研究。鉴于国内外的研究现状和发展趋势，明确了数据驱动的多元控制图失控异常模式识别、数据驱动的多元混杂数据过程控制、数据驱动的过程控制应用研究这三项研究内容。分别沿着从一元到多元、低维到高维、理论到应用的思路，从多元混杂数据的统计特性、到过程控制方法、再到控制方法的失控异常模式识别的逻辑，以及数据采集与匹配、过程控制的实施、控制结果及解释的自然逻辑，展开理论研究。依托海克斯康测量技术（青岛）有限公司的模拟数字化车间，开展数据驱动过程控制的应用研究。

面对过程数据所具有的来源多样化、数据量激增、数据类型混杂的特点，本项目“数据驱动的多元混杂数据过程控制与应用研究”对数据驱动的多元混杂数据过程控制、数据驱动的多元混杂数据异常模式识别、数据驱动的过程控制应用研究等问题进行研究。沿着统计方法和智能学习这两个研究思路，提出了基于Bonferroni’s检验以及Holm’s检验的多元混杂数据过程控制与诊断以及基于SVM的多元混杂数据过程控制与诊断这些切实可行的解决方案。将统计方法与智能学习相结合，利用PCA、PLS和小波分析进行特征提取，利用SVM和BP神经网络作为分类器，提出行之有效的多元混杂数据异常模式识别解决方案。为了推动我国质量控制水平的全面提升，实现高质量发展的战略目标，主持完成了国际标准ISO 7870-7: 2020 Control charts Part 7: Multivariate control charts和国家标准GB/T 17989.2- 2020 控制图 第2部分：常规控制图，为有效运用常规控制图和多元控制图进行统计过程控制提供了指南。