Development of a Hybrid Model for Recognition and Analysis of Significant Patterns in Process Control Charts

Correct recognition and precise classification of significant patterns in statistical process control charts is unavoidable. Because these unnatural patterns associate out of control conditions. In fact, extraction of unnatural patterns increases the sensitivity of control charts in identification of out of control states. In recent years, because of the abilities of artificial neural networks in patterns recognition, these networks have been used to discriminate unnatural patterns in Shewart control charts. In most of such studies, the misclassification error of patterns is remarkable, especially when the desired sensitivity of process is at high value. This paper proposes a hybrid model for the recognition and analysis of the basic patterns in process control charts using LVQ and MLP networks along with examining the fitted line of sample points. In the proposed model not only the misclassification error at different levels of sensitivity decreases considerably, but when basic patterns occur concurrently, the possibility of recognition of patterns and assessment of their corresponding parameters will be provided too. The efficiency and effectiveness of the model have been tested by simulated samples.