基于聚类方法的非线性轮廓线多变点问题研究

发布时间：2019-01-10 17:26

【摘要】：统计过程控制作为质量管理的重要工具之一已经越来越多地受到企业的重视,然而随着科技的发展,监控的对象已经不再是简单的一元形式。新阶段的质量控制中出现了以一个或多个自变量与响应变量之间函数关系表示的质量特性,其被称为轮廓线,对它的监控称为轮廓监控。由于轮廓监控具有处理大量数据的能力,现阶段的许多研究已不再单单局限于对产品质量的监控,而拓展到医疗、气候、经济发展等领域中。在轮廓监控方面,按照变异方式的不同可分为局部变异与全局变异,前者的变异在单体轮廓线上表现为某一段形式的偏移,后者在单体轮廓线上表现为轮廓线整体的偏移。近些年的研究多数集中于单变点领域,亦即监控过程在某一时刻发生变异,之后的过程均受该变异影响。这种假设在判断过程稳定性方面具有不错的效果,然而在真实的生产过程中,变异的类型、变异的位置以及变异的数量都是未知的,如此使得单变点的方法在实际运用过程中显得不尽如人意。因此,本文提出使用小波降噪与多阶段聚类相结合的多变点识别方法来对非线性的轮廓线过程进行监控。同时使用Matlab仿真从偏移量、变点数量、误差项方差、偏移方向等多个角度来比较本文所提方法与传统的二分法性能的优劣。通过这种多方面的比较,显示出本文所提方法在识别多变点方面无论从识别敏感度,判断稳定性还是结果可靠性上都具有相当的优势。
[Abstract]:As one of the important tools of quality management, statistical process control has been paid more and more attention by enterprises. However, with the development of science and technology, the object of monitoring is no longer a simple unitary form. In the new stage of quality control, the quality characteristics are represented by the functional relationship between one or more independent variables and the response variables, which are called contour lines, and the monitoring of them is called contour monitoring. Because contour monitoring has the ability to deal with a large number of data, many researches at present are no longer confined to the monitoring of product quality, but have been extended to the fields of medical treatment, climate, economic development and so on. In the aspect of contour monitoring, local variation and global variation can be divided into local variation and global variation according to the variation mode. The variation of the former is an offset of a certain section on the single contour line, and the latter is the overall deviation of the contour line on the single contour line. In recent years, most studies have focused on the field of single change point, that is, the monitoring process changes at a certain time, and then the process is affected by the variation. This assumption has a good effect on judging process stability, but in real production, the type of variation, the location of variation, and the number of variations are unknown. In this way, the method of single change point is not satisfactory in the process of practical application. Therefore, this paper proposes a multi-point recognition method which combines wavelet denoising with multi-stage clustering to monitor the nonlinear contour process. At the same time, Matlab simulation is used to compare the performance of the proposed method with that of the traditional dichotomy from several angles, such as offset, number of change points, variance of error terms, offset direction and so on. Through this multi-aspect comparison, it is shown that the method proposed in this paper has considerable advantages in identifying multivariate points in terms of sensitivity, stability and reliability of results.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F272

【参考文献】