(X)OVER-BAR CONTROL CHART PATTERN IDENTIFICATION THROUGH EFFICIENT OFF-LINE NEURAL-NETWORK TRAINING

Back-propagation pattern recognizers (BPPR) are proposed to identify u nnatural patterns exhibited on Shewhart control charts. These unnatura l patterns, such as cycles and trends, can provide valuable informatio n for real-time process control. In a computer-integrated manufacturin g environment, the operator need not routinely monitor the control cha rt but, rather, can be alerted to patterns by a computer signal genera ted by the propose algorithm. In this paper, an off-line analysis is p erformed to investigate the training and learning speed of these BPPRs on simulated xBAR data. The best configuration of the network is furt her tested to demonstrate the classification capability of the propose d BPPR.