An approximate summarization method of process data for acquiring knowledge to improve product quality

This paper describes a machine learning approach for a manufacturing databa se. The method is presented in the Nb-Ti superconducting wire domain. A Nb- Ti superconducting wire is produced by iterating the drawing and heat treat ment operations. The purpose is to obtain approximate summarization of proc ess data that describes how a production schedule can be improved for bette r product quality. The method consists of the following steps: First, der n e a ranking function for a production schedule. Then, generate `positive' a nd `negative' instances for improving a production schedule by comparing a pair of schedules and their ranking values in the database. Using a machine learning technique, called `ID3', a `modification patterns' are obtained t hat generalize the data for better production quality. The final step is to extract approximate information from the induced patterns, which is both d esirable for easier understanding by human experts and necessary to avoid b eing too much influenced by excessive details or disturbances. Two criteria are proposed, correctness and applicability indices, for this approximatio n.