|
Activity Number:
|
102
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2008 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Quality and Productivity
|
| Abstract - #302646 |
|
Title:
|
A Residual-Based Hotelling's T^2 Chart
|
|
Author(s):
|
Soren Bisgaard+ and Xuan Huang*+
|
|
Companies:
|
University of Massachusetts Amherst and University of Massachusetts Amherst
|
|
Address:
|
, , , Isenberg School of Management, Amherst, MA, 01003,
|
|
Keywords:
|
Process monitoring ; Statistical process control ; Quality control ; Multivariate time series ; Average run length
|
|
Abstract:
|
The application of Hotelling's T^2 for the simultaneous monitoring of several quality characteristics assumes temporal independence. However, industrial data is often autocorrelated. For applications where the data stream follows a multivariate time series model, we propose a special cause control chart based on applying Hotelling's T^2 to the residuals. We provide detailed evaluations of the performance of the proposed method based on the Average Run Length (ARL) for a group of first order vector autoregressive models. We conclude that the residual based method in most cases perform favorably relative to using a Hotelling's T^2 chart directly on the unfiltered observations. We also demonstrate the method when applied to data from an industrial process.
|
