Categorization is the procedure of determining set membership based on either necessary or statistically suggestive conditions for membership. This procedure lies at the heart of automated metallurgical failure analysis, controlling the accuracy of the final conclusion. This article examines the tradeoff between the number of questions posed by the computer during data collection and the certainty of the final decision. After a brief overview of failure analysis decision making, a model of categorization is proposed which is derived from Bayes’ theorem that asks questions in order of relevance and stops when an adequate level of certainty is achieved. This eliminates irrelevant questions without significantly compromising the accuracy of the final conclusion. The model has been implemented as part of an artificial intelligence computer program.

This content is only available via PDF.
You do not currently have access to this content.