Diagnostic Feature Extraction From Stamping Tonnage Signals Based on Design of Experiments

Abstract

Diagnostic feature extraction with consideration of interactions between variables is very important, but has been neglected in most diagnostic research. In this paper, a new feature extraction methodology is developed to consider variable interactions by using a fractional factorial design of experiments (DOE). In this methodology, features are extracted by using principal component analysis (PCA) to represent variation patterns of tonnage signals. Regression analyses are performed to model the relationship between features and process variables. Hierarchical classifiers and the cross-validation method are used for root-cause determination and diagnostic performance evaluation. A real-world example is used to illustrate the new methodology. [S1087-1357(00)00302-6]