Automating the Diagnosis and Rectification of Deflection Yoke Production Using Hybrid Knowledge Acquisition and Case-Based Reasoning

Abstract

Deflection yoke (DY) is one of the main components of the color display tube (CDT) that determines the image quality of a computer monitor. Once a DY anomaly is found during production, the remedy process is performed in two steps: identifying the type of anomaly from the observed problem pattern and adjusting manufacturing process parameters to rectify it. To support this process, we introduce a knowledge-based system using a hybrid knowledge acquisition technique and case-based reasoning. The initial phase of the knowledge acquisition employs a systematic and quantitative data processing including stepwise regression and an inductive learning algorithm. This automated expertise elicitation produces strategies, which are represented by decision trees or if-then rules, to specify DY anomalies from display patterns. The strategies are then refined by introducing human expertise. The knowledge acquisition process was designed to support for this cognitive cooperation. For coordinating the process parameters to remedy the specified anomalies, a case-based reasoning is utilized. The laboratory and field test proved that the developed knowledge-based system could produce highly effective decisions for the process control in DY production.