A Method for Reasoning by Analogy in Failure Analysis

Abstract

An artificial intelligence computer program is being developed for automating the analysis of metallurgical failures. The purpose of this expert system is for recognizing modes of failure like stress corrosion and hydrogen embrittlement. We have previously described a method for recognizing mode of failure based on a form of diagnosis or statistical categorization. In this paper, an alternative method is developed for recognizing mode of failure as a means of confirming or disconfirming the conclusions derived by categorizing. This method is a form of reasoning by analogy that recognizes when the current case is similar to some previously encountered case and uses this information to draw conclusions. The expert system is “programmed” by providing a set of correctly solved examples rather than IF-THEN rules. Accuracy of analogical reasoning is found to increase with number of stored examples, and in recognizing modes of failure an error rate as low as 20.7 percent is observed to occur.