Collaborative Research: Blind Discovery of Variation Sources for Visualization by Multidisciplinary Teams

系统地减少制造差异和提高质量的计划(通常被称为六西格玛计划)现在已经在工业中站稳了脚跟。然而，绝大多数六西格玛分析工具是几十年前设计的，用于结构相对简单的有限数据量。这项研究将开发一种针对现代制造操作中发现的高维数据结构而设计的六西格玛差异缩减的知识发现方法。与为已知变异源建立先验模型，然后在数据中寻找特定的预建模模式的方法不同，本研究的目标是盲目地发现变化模式及其来源的性质，仅基于数据样本，而不是预建模。每个识别的模式的交互式图形可视化将使用户能够可视化变化的根本原因。为了实现这一点，这项研究将开发一种表示变化模式的范例，该模式包含各种数据结构中的线性和非线性现象。这项研究还将开发盲目识别模式的算法，尽可能地提高准确性、稳健性和自动化程度。如果成功，这项研究将使减少差异的方法现代化，以跟上测量和信息技术的进步。虽然算法必然很复杂，但自动化，加上将开发的可视化结果的方法，将创造出易于使用和广泛适用的工具。这些特征是传统的六西格玛标志，将促进方法的广泛传播和采用，并使多学科合作者团队(例如，具有不同背景的操作员、工程师、统计学家和管理人员)能够使用该方法。将把成果纳入面向本科生和博士生的六西格玛和数据挖掘课程，并向专业硕士课程中不同族裔和不同技术的工程师和管理人员提供这些课程，从而进一步加强传播。

Programs for systematically reducing manufacturing variation and improving quality (often called six-sigma programs) are now firmly established in industry. However, the vast majority of six-sigma analysis tools were designed decades ago for use with limited amounts of data with relatively simple structure. This research will develop a knowledge discovery methodology for six-sigma variation reduction that is designed for the high-dimensional data structures found in modern manufacturing operations. In contrast to methods in which one constructs prior models for known variation sources and then looks for those specific premodeled patterns in the data, the goal of this research is to blindly discover the nature of the variation patterns and their sources, based solely on a sample of data, with no premodeling. Interactive graphical visualizations of each identified pattern will enable users to visualize the root causes of variation. To accomplish this, this research will develop a paradigm for representing variation patterns that encompasses linear and nonlinear phenomena in a variety of data structures. The research will also develop algorithms for blindly identifying the patterns with as much accuracy, robustness, and automation as possible. If successful, this research will modernize variation reduction methods to keep pace with advances in measurement and information technology. Although the algorithms will necessarily be complex, the automation, coupled with methods that will be developed for visualizing the results, will create tools that are easy-to-use and widely applicable. These characteristics, which are traditional six-sigma hallmarks, will facilitate broad dissemination and adoption of the methodology and enable its use by multidisciplinary teams of collaborators (e.g., operators, engineers, statisticians, and managers with varying backgrounds). Dissemination will be further enhanced by integrating the results into six-sigma and data mining courses offered to undergraduate and Ph.D. students and to an ethnically and technically diverse spectrum of engineers and managers in professional masters' courses.