Online learning of the cause-and-effect knowledge of a manufacturing process

This paper deals with the intelligent learning of the cause-and-effect knowledge of a manufacturing process in online mode. This knowledge discovery problem is characterized as online learning where the knowledge is gradually found using the instances periodically obtained from the part processing of the process. We develop a new decision tree learning method called 'Statistical Batch based Decision tree Learning' (SBDL). To deal with large number of instances collected from the process, the concept of batch-based learning is introduced. A two-phased fitness test is also developed for measuring the fitness of the decision tree, thereby detecting the update point in time of the decision tree. The performance of SBDL has been verified with a real instance set collected from a Korean TFT-LCD manufacturing company.