Phase I Monitoring of Univariate Processes Using Hierarchical Clustering: A Two-Step Approach with Unique Measures of Dissimilarity

• Activity Number: 214 - Contributed Poster Presentations: Quality and Productivity Section
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Quality and Productivity Section
• Author(s): Bryce Whitehead* and Austin Brown
• Companies: University of Northern Colorado and Kennesaw State University
• Keywords: Statistical Process Control; Mixture distribution; Hierarchical clustering

Abstract

When implementing a statistical process control (SPC) charting scheme, it is important for chart operators to ensure the monitored process is in a state of statistical control in Phase I prior to on-line monitoring in Phase II. In many traditional univariate control charting schemes, it is assumed the monitored process follows a normal distribution. Literature has shown that if this assumption is not met, then both the in and out-of-control average run length can be adversely affected in Phase II. For example, a process’ distribution may be a mixture of two or more distributions depending on the nature of the quality characteristic. Several studies have examined the influence mixture distributions have upon control chart performance, but there is scant information in the literature regarding how to handle or detect such situations in Phase I. This study proposes use of two dissimilarity metrics based on the area and location of a Phase I sample in hierarchical clustering as a means of monitoring the possible mixture of both the scale and location parameters. The efficacy of this technique is examined, and recommendations are given.