Some improvements to Taguchi analysis
Abstract
In most quality improvement circles, the concept of continuous improvement is paramount, and yet the designed experiment, or a Taguchi experiment is often applied in a 'one-shot-only' fashion. This thesis examines the Taguchi method and philosophy, and suggests some improvements in the design and analysis, following a more classical statistical approach. Various methods and criterion are offered in an attempt to synthesize all information available in order to select variables and the appropriate levels for a subsequent experiment. The bootstrap is used in order to assess the uncertainty arising in estimating unknown parameters, and the effect of using estimated quantities is also considered. It is hoped that the next experiment will deliver output with a lower mean squared error in the response relative to a desired target level. Lastly, some areas for future research are discussed.
Degree
Ph.D.
Advisors
Kuczek, Purdue University.
Subject Area
Statistics
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.