Failure Amplification Method: An Information Maximization Approach to Categorical Response Optimization

Abstract

Categorical data arise quite often in industrial experiments because of an expensive or inadequate measurement system for obtaining continuous data. When the failure probability/defect rate is small, experiments with categorical data provide little information regarding the effect of factors of interest and generally are not useful for product/process optimization. We propose an engineering-statistical framework for categorical response optimization that overcomes the inherent problems associated with categorical data. The basic idea is to select a factor that has a known effect on the response and use it to amplify the failure probability so as to maximize the information in the experiment. New modeling and optimization methods are developed. The method is illustrated with two real experiments.