复杂制造过程中轮廓数据监控方法研究

Statistical process control is widely used to monitor the product (or process) quality characteristics in manufacturing process. In complex manufacturing process, the product (or process) quality characteristics include not only the product dimension, but also the product profile. Traditional statistical process control approaches assume that the quality characteristics follow a special distribution. However, the product profile, which can be characterized by a function relationship, obviously cannot meet this assumption. Hence, some control methods should be developed to monitor the product (or process) profiles. According to the different typical profile changes in complex manufacturing processes, this project aims to study the profile monitoring approaches, which involves the following research contents: (1) the effective monitoring and diagnosing of local profile changes when the change magnitude is small; (2) the detection of several types of anticipated profile shape changes, and the directed process monitoring of anticipated profile changes; (3) the simultaneously monitoring of profile mean and correlation by considering the within-profile correlation. The study of profile monitoring is an important complement of statistical process control. This research has demonstrated the potential in the aspect of monitoring the profile data for empirical study in the complex manufacturing process or in the service fields such as public health care and telecommunication industry.

统计过程控制理论广泛应用于制造过程中产品（或过程）质量特性的监控。在复杂制造过程中，某些产品（或过程）质量特性不仅包括产品尺寸，还包括产品形状。传统的统计过程控制方法通常假设质量特性服从某种一元或多元分布。然而，满足某种函数关系的形状轮廓显然不满足此假设。因此，需要研究产品（或过程）轮廓数据的监控方法。本项目针对复杂制造过程中不同轮廓变异情况，分别研究轮廓监控方法，包括：（1）针对轮廓局部发生小偏移变形的情况，研究有效的轮廓局部变形监控与诊断方法；（2）研究轮廓发生特定变异情况下，几种特定变异的轮廓监控方法；在此基础上，研究一般类型特定变异的轮廓监控方法；（3）考虑轮廓内存在相关性的特点，研究同时监测轮廓均值和相关性的轮廓监控方法。本项目研究成果对于拓展轮廓数据的统计过程控制方法研究具有重要的理论价值；在实践上可用于复杂制造过程、医疗保健和电讯业等服务过程中轮廓数据的监测分析。

在国家自然科学基金青年项目的资助下，项目组针对复杂制造过程中产品或过程轮廓数据和质量特性数据的特点，主要完成了如下几个方面的工作：（1）采用相加模型和广义线性模型对特定变异轮廓进行了建模，并基于此模型构建了有效的轮廓特定变异EWMA、CUSUM、带有附加运行准则和区间控制图等监控方法，之后通过仿真技术分析了所提方法的性能。（2）对多元自相关数据和二元二项自相关数据的过程监控方法进行了研究；在此基础上对轮廓内部存在自相关的情况，分析了自相关参数估计对第二阶段内部自相关轮廓控制图的影响。（3）针对轮廓发生局部变形的特点，提出了基于二维空间的修正豪斯多夫算法的非参数轮廓诊断方法；当轮廓输出变量为离散变量时，提出了解释变量分别为固定取值和非固定取值情况下的轮廓变点诊断方法。（4）针对产品质量特性数据，提出了产品特性数据服从Weibull分布且存在截尾情况下的数据分析方法；针对产品质量数据高维非平衡性，提出了非平衡数据关键质量特性识别方法。. 到目前为止，项目组共发表（含录用）论文15篇，其中SCI检索论文10篇。培养博士研究生1名，硕士研究生1名，正在培养硕士研究生1名，参加国内外学术交流活动7次。