Recognition of unnatural patterns in process control charts through combining two types of neural networks

Unnatural patterns in control charts associate a set of assignable causes in production process. Therefore, effective discrimination of these patterns has special importance in on-line statistical process control. In resent years, artificial neural networks because of their abilities in patterns recognition have been used to detect control charts patterns. The correct and precise recognition in a real-time is difficult, because control chart patterns are distorted by common cause variations that occur naturally in the manufacturing processes. Nearly the most of investigations based neural networks solely have emphasized the control chart patterns recognition and have not considered extraction of detailed information that is important for analysis of assignable causes. Moreover, some of the patterns simulator functions do not represent completely the real world behavior. This study presents a hybrid model, including two types of neural networks. The proposed model can detect and discriminate single and concurrent patterns, determine the major corresponding parameters and estimate the starting point of unnatural patterns. This model introduces a procedure for patterns simulation in variable and attribute qualitative characteristics control charts. In addition, this research improves and develops unnatural patterns generator functions. Numerical results indicate that the proposed model performs suitably and effectively.