Change Detection for Multi-Stage Multivariate Data

• Activity Number: 643 - Detection of Changes and Structural Breaks in Business and Industrial Data Streams
• Type: Topic Contributed
• Date/Time: Thursday, August 1, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Quality and Productivity Section
• Author(s): Emmanuel Yashchin*
• Companies: IBM Research
• Keywords: Detection; Monitoring; Quality Control; Search Engine; Statistical Process Control

Abstract

We consider early warning systems (EWS) for monitoring multi-stage data, in which downstream variables undergo changes associated with upstream process stages. In such applications, timeslides are typically useful for both monitoring and diagnostics. A timeslide is a set of observations that is ordered in accordance with product passage time through a given stage. In this paper, we discuss using timeslides in multivariate methodology for detecting unfavorable changes in downstream variables. We illustrate the methodology using examples from electronic industry.