Defect Cause Search Support System Using Ontology and Bayesian Network in Liquid Crystal Display Manufacturing Process

Abstract

This paper describes a new defect cause search support system by combining Bayesian network and ontology to use the inherent production theory and the operation knowledge in the manufacturing process. The internal parameters, that are not measured as the actual data but can be calculated by mathematical model based on the production theory, often cause the truth of defects. In this study, these parameters information is used for the probabilistic inference of the defect cause by Bayesian network. The ontology is used for data dimension reduction, because too many dimensions of data deteriorates the estimation accuracy. Moreover, the calculation method of the similarity degree between the concepts in the ontology is used to search the new information about the production theory and the operation knowledge. This system was applied to the actual defect analysis in the liquid crystal display manufacturing process. As the result, F-measure that means the accuracy in Bayesian network has increased by 0.4905 drastically and the analysis time can be shortened to about 1/3.