Ensemble ANN-Based Recognizers to Improve Classification of X-bar Control Chart Patterns

Many of the previous research on the control chart pattern recognition were related to fully developed patterns. However, in practice, the process data will appear as a continuous stream of partially developed patterns. Such developing patterns are difficult to recognize since their structure are normally vague and dynamic. This study investigated the merit of a generalized single recognizer (all-class-one-network (ACON), a committee of specialized recognizers (one-class-one network, OCON) and the ensemble of ACON and OCON recognizers. These recognizers were embedded into a monitoring framework to enable on-line recognition. The performance of the schemes was evaluated based on percentage correct classification. The findings suggest that the ensemble of ACON and OCON recognizers with simple summation could significantly improve its discrimination capability. It is concluded that the strategy to configure and consolidate multiple recognizers is very important to achieve good classification performance.