Defect Cause Search Support System Using Manufacturing Information Ontology and Bayesian Network in LCD Manufacturing Process
Kouki Hamamoto, Akira Kitamura, Satoshi Taguchi, Takahiro Shoji, Hiroki Matsuno, Shingo Watanabe
Usually, in the manufacturing process, Quality control methods such as the QC procedure have been used for the defect cause analysis. However, in these methods, it is difficult to identify complicated defect causes in which multiple factors are intricately intertwined in high-accuracy and short time. Moreover, the manufacturing theory and the operation knowledge belonging in engineers aren't used fully. Therefore, this paper proposes the new defect cause search support system by the cause estimation model using Bayesian network and manufacturing theory and knowledge in Ontology. In this system, the deeper causes are searched by the model that is constructed from the actual data and the theory information. The discretization method using decision tree and the feature selection method using Ontology and SVM are used as preprocessing for the model construction. Authors have applied this system to LCD manufacturing process. As the result, estimation accuracy of the model has improved 14.78% in the accuracy rate and analysis time has been shortened to about 1/4. It is possible that the defect causes are identified from various standpoints such as manufacturing theory in high-accuracy and short time by this system.